arX

iv:1

604.

0176

8v1

[as

tro-

ph.G

A]

6 A

pr 2

016

Accepted for publication in ApJPreprint typeset using LATEX style emulateapj v. 01/23/15

THE GLOBULAR CLUSTER SYSTEM OF THE COMA cD GALAXY NGC4874 FROM HUBBLE SPACETELESCOPE ACS AND WFC3/IR IMAGING*

Hyejeon Cho1, John P. Blakeslee2, Ana L. Chies-Santos3,4, M. James Jee1, Joseph B. Jensen5,Eric W. Peng6,7 & Young-Wook Lee1

Accepted for publication in ApJ

ABSTRACT

We present new Hubble Space Telescope (HST ) optical and near-infrared (NIR) photometry of the richglobular cluster (GC) system of NGC4874, the cD galaxy in the core of the Coma cluster (Abell 1656).NGC4874 was observed with the HST Advanced Camera for Surveys in the F475W (g475) and F814W(I814) passbands and the Wide Field Camera 3 IR Channel in F160W (H160). The GCs in this fieldexhibit a bimodal optical color distribution with more than half of the GCs falling on the red side atg475−I814 > 1. Bimodality is also present, though less conspicuously, in the optical-NIR I814−H160

color. Consistent with past work, we find evidence for nonlinearity in the g475−I814 versus I814−H160

color-color relation. Our results thus underscore the need for understanding the detailed form of thecolor-metallicity relations in interpreting observational data on GC bimodality. We also find a verystrong color-magnitude trend, or “blue tilt,” for the blue component of the optical color distributionof the NGC4874 GC system. A similarly strong trend is present for the overall mean I814−H160

color as a function of magnitude; for M814 < −10 mag, these trends imply a steep mass-metallicityscaling with Z ∝ M1.4±0.4

GC , but the scaling is not a simple power law and becomes much weaker atlower masses. As in other similar systems, the spatial distribution of the blue GCs is more extendedthan that of the red GCs, partly because of blue GCs associated with surrounding cluster galaxies.In addition, the center of the GC system is displaced by 4 ± 1 kpc towards the southwest from theluminosity center of NGC 4874, in the direction of NGC4872. Finally, we remark on a dwarf ellipticalgalaxy with a noticeably asymmetrical GC distribution. Interestingly, this dwarf has a velocity ofnearly −3000 km s−1 with respect to NGC4874; we suggest it is on its first infall into the cluster coreand is undergoing stripping of its GC system by the cluster potential.

Keywords: galaxies: elliptical and lenticular, cD — galaxies: individual (NGC 4874) — galaxies:clusters: individual (Coma) — galaxies: star clusters — globular clusters: general

1. INTRODUCTION

All large galaxies possess globular cluster (GCs) popu-lations, or systems, comprising hundreds or thousands ofindividual GCs. They are often used as discrete tracers ofgalaxy assembly, especially at the outer regions of largegalaxies where they can be more easily observed than thefaint integrated galaxy light. Moreover, they provide in-formation on the various processes and progenitors thatare present in the different phases of the build up of early-type galaxies (Zaritsky et al. 2014), which are the domi-

* Based on observations with the NASA/ESA Hubble SpaceTelescope, obtained from the Space Telescope Science Institute(STScI), which is operated by AURA, Inc., under NASA contractNAS 5-26555. These observations are associated with program#11712.

1 Department of Astronomy and Center for Galaxy Evo-lution Research, Yonsei University, Seoul 03722, Korea;hyejeon@yonsei.ac.kr; ywlee2@yonsei.ac.kr

2 NRC Herzberg Astronomy & Astrophysics, Victoria, BCV9E 2E7, Canada; john.blakeslee@nrc-cnrc.gc.ca

3 Departamento de Astronomia, Instituto de Fısica, UFRGS,Porto Alegre, R.S. 91501-970, Brazil

4 Departamento de Astronomia, IAGCA, Universidade de SaoPaulo, 05508-900 Sao Paulo, SP, Brazil

5 Department of Physics, Utah Valley University, Orem, Utah84058, USA

6 Department of Astronomy, Peking University, Beijing100871, China

7 Kavli Institute for Astronomy and Astrophysics, Peking Uni-versity, Beijing 100871, China

nant structures at the centers of galaxy clusters (Dressler1980). Interestingly, the number of GCs, or the totalmass of the GC system, appears to be related to the massof the dark matter halo of the galaxy or galaxy cluster(Blakeslee et al. 1997; Blakeslee 1999; Bekki et al. 2008;Spitler & Forbes 2009; Alamo-Martınez et al. 2013; Hud-son et al. 2014). However, even at a given mass, thereis significant galaxy-to-galaxy scatter in the number andother properties of the GC systems, and this scatter islikely the result of environmental effects and stochasticvariations in the galaxy formation histories (e.g., Penget al. 2008; Harris et al. 2013). GC formation requiresvery high star formation rates, and thus the major starformation episodes and assembly histories of early-typegalaxies can be traced by the observed properties of theirGC systems, such as their colors, metallicities, and spa-tial distributions (e.g., Brodie & Strader 2006; Peng et al.2006; Forte et al. 2014).Since nearly all the GCs surrounding massive galaxies

are old, metallicity is the main stellar population variableamong individual GCs within a GC system. However, atpresent the acquisition of large samples of spectroscopicmetallicities for large numbers of GC systems is impracti-cal. With the exception of a few nearby early-type galax-ies such as NGC 5128 (e.g., Beasley et al. 2008; Woodleyet al. 2010), the majority of the Lick index-type spec-troscopic studies using 10m class telescopes have beencarried out on samples of ∼ 20 to 50 GCs. One recent

2 Cho et al.

major effort on this front is the SLUGGS survey (Brodieet al. 2014), which uses the calcium II triplet (CaT) in-dex as a metallicity proxy for over 1000 GCs within asample of 25 GC systems. CaT index distributions havebeen investigated by Usher et al. (2012), and a clear caseof bimodality was presented by Brodie et al. (2012) forthe edge-on S0 galaxy NGC3115.Obtaining spectra with sufficiently high signal-to-noise

ratios to measure accurate metallicities for large samplesof extragalactic GCs is difficult; obtaining large samplesof high-quality photometric colors is much simpler. Opti-cal color distributions for GC systems of massive galaxiesare generally found to be bimodal (e.g., Peng et al. 2006);however, this bimodality does not always hold when con-sidering other color combinations including near-infrared(NIR; Blakeslee et al. 2012; Chies-Santos et al. 2012) orUV bands (Yoon et al. 2011a,b). Cantiello & Blakeslee(2007; see also Puzia et al. 2002) have shown that opti-cal/NIR colors are better metallicity proxies than purelyoptical colors. Optical wavelengths in old stellar systemsare sensitive to the red giant branch, but also to starson the horizontal branch (HB) and near the main se-quence turn-off point, and therefore are degenerate in ageand metallicity. The NIR wavelength range is dominatedby red giant branch stars, so the optical-NIR colors aretherefore mainly sensitive to metallicity. Moreover, Yoonet al. (2006) have shown that non-linear color-metallicityrelations, possibly related to a sharp transition in theHB morphology at a certain metallicity, can transforma unimodal metallicity distribution into bimodal opticalcolor distributions (see also Richtler 2006). Despite bi-modal metallicity distributions found for galaxies suchas NGC 3115 and the Milky Way, color-color nonlineari-ties are present to a certain degree (Cantiello et al. 2014;Vanderbeke et al. 2014) and are not yet fully under-stood. In a study of stacked GC spectra around theCaT region for 10 galaxies, Usher et al. (2015) foundgalaxy-to-galaxy variations in the CaT-color relations,implying that different types of galaxies require differ-ent color-metallicity transformations for estimating GCmetallicities from photometric data.The situation is even more complex. The bimodality

in the optical color distributions of GCs around mas-sive galaxies varies in both the relative proportions ofblue and red GCs and in the mean colors of these twocomponents. Even within a given galaxy there are vari-ations. For instance, the blue GCs of certain massivegalaxies are observed to have redder colors at brightermagnitudes. This color-luminosity relation, or “blue tilt”(Harris et al. 2006; Mieske et al. 2006; Strader et al.2006; Wehner et al. 2008) has been suggested to be dueto self-enrichment (e.g., Strader & Smith 2008; Bailin& Harris 2009). The blue tilt only becomes significantfor GCs with masses above ∼ 106M⊙ but this effect hasimportant implications for color distribution studies asthe location of the blue and red peaks will vary with themagnitude range of the GCs considered.The Coma cluster of galaxies is a truly massive and

rich galaxy cluster at a mean redshift of z = 0.024 (Col-less & Dunn 1996), corresponding to a distance of about100Mpc (for h = 0.7). Its virial mass of 2.7 × 1015 M⊙

(Kubo et al. 2007) is roughly four times more massivethan the Virgo cluster (see Carter et al. 2008; Durrell

et al. 2014). As the anchor of comparison for study-ing properties of both galaxies and clusters between thenearby and distant universe, Coma presents an attractiveopportunity for detailed studies of GCs in a dense clus-ter environment (e.g., Harris 1987, Harris et al. 2009;Blakeslee & Tonry 1995; Blakeslee et al. 1997; Marın-Franch & Aparicio 2002). Analyzing the data from theHST/ACS Coma Cluster Treasury Survey (hereafter AC-SCCS; Carter et al. 2008), Peng et al. (2011) discovereda population of intracluster globular clusters (IGCs) inComa that did not appear to be associated with anygalaxy. This Coma IGC population was estimated tomake up ∼ 30–45% of all GCs in the cluster core andpresents a bimodal color distribution with blue GCsgreatly outnumbering red ones. Much of the remain-ing portion of Coma’s core GC system belongs to its cDgalaxy NGC4874, which also has a bimodal color dis-tribution, but with a blue population that is somewhatredder than the blue IGCs.This paper presents an analysis of new, significantly

deeper ACS optical data than was obtained by theACSCCS, with the addition of new high resolutionHST/WFC3 NIR photometry of the GC system sur-rounding NGC4874. We study the optical and optical-NIR color distributions, the nonlinear behavior in thecolor–color relations as well as color–magnitude trends inthe ACS/WFC F475W, F814W and WFC3/IR F160Wbandpass combinations. The spatial distribution of theGC system is also explored within the wider ACS/WFCfield of view. For consistency with the ACSCCS stud-ies (e.g., Carter et al. 2008; Peng et al. 2011) we adoptthroughout this paper a distance of 100 Mpc to Coma,giving a distance modulus of (m−M) = 35.0 mag.At this distance, 1′′ corresponds to a physical scale of0.48 kpc.

2. OBSERVATIONAL DATA SETS

As part of HST program GO-11711, we imagedNGC4874 with the Advanced Camera for Surveys WideField Channel (ACS/WFC) for four orbits in F814W(I814) and one orbit in F475W (g475); to this, we addedadditional imaging in g475 from GO-10861. The field ofview of the ACS/WFC is approximately 3.′37× 3.′37. Theexposures were dithered to improve bad pixel rejectionand to fill in the 2.′′5 gap between the two ACS/WFCdetectors. Following the standard pipeline processing atthe Space Telescope Science Institute’s Mikulski Archivefor Space Telescopes (MAST), we used the stand-aloneversion of the empirical pixel-based charge-transfer effi-ciency (CTE) correction algorithm of Anderson & Bedin(2010) on each of the individual calibrated “flt” expo-sures to remove the CTE trails from the ACS data. Thecalibrated, CTE-corrected exposures were then processedwith Apsis (Blakeslee et al. 2003) to produce geometri-cally corrected, cosmic-ray rejected stacked images witha final pixel scale of 0.′′05 pix−1.Figure 1 (a) shows our ACS/WFC F814W image of the

NGC4874 field, along with the designations and morpho-logical classifications from the NASA/IPAC Extragalac-tic Database (NED)8 for eight bright galaxies (includingNGC4874 itself).

8 http://ned.ipac.caltech.edu

NGC 4874 Globular Clusters 3

(a)

N E

NGC 4871 (SAB0/a)

NGC 4872 (SB0)

NGC 4874 (cD)

NGC 4873 (SA0) PGC 44644 (SA0)

PGC 44651 (S0)

PGC 44652 (S0)

PGC 44636 (SA0/a)

(b)

N E

(c)

N E

(d)

N E40′′ = 19.2 kpc

Figure 1. Stacked HST ACS/WFC F814W image of the field roughly centered on NGC4874. The size of the field is 3.′52×3.′56. (a) Thedrizzled science image, shown at the observed orientation. Bright extended galaxies in the field are labeled and their morphological typesfrom NED are shown in parentheses. (b) The sum of ELLIPSE-generated isophotal models for NGC4874 and nine surrounding galaxies.(c) The sum of galaxy isophotal models and a large-scale residual map constructed using SExtractor. (d) The final “residual image” usedfor the object detection with the isophotal models and SExtractor background subtracted.

We also observed NGC4874 with the Wide Field Cam-era 3 IR Channel (WFC3/IR) in parallel for six addi-tional orbits of GO-11711, with four of the orbits in thelongest wavelength F160W (H160) bandpass, during pri-mary ACS/WFC observations of the neighboring Comagiant elliptical NGC 4889. The WFC3/IR focal plane ar-ray consists of a single detector with a field of view of2.′27× 2.′05. The calibrated WFC3/IR H160 exposureswere retrieved from STScI/MAST and combined into a

final geometrically corrected image using the MultiDriz-zle (Koekemoer et al. 2003; Fruchter et al. 2009) taskin the PyRAF/STSDAS package9. As in Blakeslee et al.(2012), we used an output pixel scale of 0.′′1 pix−1, whichis twice that of ACS/WFC. Table 1 summarizes the ob-servational details of our imaging data; note that the two

9 PyRAF and STSDAS are products of the Space Telescope Sci-ence Institute, operated by AURA for NASA.

4 Cho et al.

Table 1Observational Details of the Data Sets

Program Dataset Instrument/ Bandpass Exp. Time m1a Magnitude

ID Detector (sec) (mag) Symbol

11711 JB2I01010 ACS/WFC F475W 2394.0 26.056 g47510861 J9TY19040 ACS/WFC F475W 2677.0 26.045 g47511711 JB2I01020 ACS/WFC F814W 10425.0 25.947 I81411711 IB2I02040 WFC3/IR F160W 10790.8 25.946 H160

a Photometric zeropoints represent the magnitudes on the AB system corresponding toone count per second.

sets of F475W data from the two different programs werecombined by Apsis into a single stacked image.We corrected for Galactic extinction toward NGC4874

assuming E(B−V ) = 0.0091 mag (Schlegel et al. 1998)and the revised ACS/WFC and WFC3/IR extinctioncoefficients (for RV =3.1) from Schlafly & Finkbeiner(2011); the resulting corrections were small, amountingto 0.030, 0.014, and 0.005 mag in g475, I814, and H160,respectively. When we derived K-corrections for 12 Gyrmodel spectral energy distributions, which are redshiftedto the NGC4874 distance (Benıtez 2000), with [Fe/H]= −1.7 and −0.7 (Bruzual & Charlot 2003; C. Chung,private communication), corresponding to blue and redpeak GCs, the average corrections are 0.05, 0.00, and−0.02 mag for g475, I814, and H160, respectively. Sinceit is uncertain how good the evolutionary stellar popu-lation synthesis models are at NIR wavelengths and theestimated K-corrections are small but model-dependent,we have not applied them to our magnitudes and colorsfor GC candidates.In this paper, we calibrate the ACS photometry to

the AB system following Bohlin (2012) and adopting thetime-variable zero points from the online ACS ZeropointsCalculator10. The WFC3 photometry is calibrated usingthe AB zero points from the online WFC3 zero pointtables11 (06 March 2012 revision). For reference, theadopted zero points are provided in Table 1; in the caseof F475W, we used an exposure time-weighted averageof the zero points for the two different observations.

3. PHOTOMETRIC ANALYSIS

3.1. Galaxy and Background Subtraction

In order to detect point-like objects embedded in theextended galaxy halo light, we first removed the smoothgalaxy light profiles from the final combined images. Weconstructed elliptical isophotal models for each of thebright galaxies in each of the stacked bandpass imagesusing the IRAF/STSDAS tasks ELLIPSE and BMODEL,which use the fitting algorithm and the uncertainty es-timation method described by Jedrzejewski (1987) andBusko (1996). We started by making an initial model(improved with later iterations) of the brightest galaxy(NGC4874), then progressed by modeling the othergalaxies in order of their luminosity. When runningELLIPSE, we first masked bright foreground stars, badpixels, and any bright galaxies in the field except for thegalaxy being fitted; then we modeled the isophotes ofthe galaxy light distribution. Using the isophotal pa-rameters from ELLIPSE, we then build a smooth galaxy

10 http://www.stsci.edu/hst/acs/analysis/zeropoints/zpt.py11 http://www.stsci.edu/hst/wfc3/phot_zp_lbn

model with BMODEL. After subtracting the model from theoriginal image, we fitted isophotes of the next brightestgalaxy and subtracted this isophotal model as well. Werepeated this process until we had subtracted ten galax-ies in the ACS/WFC image and four galaxies (NGC 4874itself and three surrounding galaxies) in the WFC3/IRimage. As mentioned above, it was necessary to modelthe galaxies iteratively in order to achieve the cleanestmodel subtractions (e.g., Alamo-Martınez et al. 2013).After subtracting the elliptical isophotal models, we

modeled the residual background using SExtractor(Bertin & Arnouts 1996) to fit a two-dimensional bicu-bic spline with the parameters BACK SIZE = 32 andBACK FILTERSIZE = 3. This removes residual structureon scales much larger than the full width half maximum(FWHM) of the point spread function (PSF), and thusdoes not detrimentally affect the point source photome-try (see Jordan et al. 2004). We note that subtractionof the isophotal model generated by the BMODEL tasksometimes results in a noticeable discontinuity in sur-face brightness at the “edge” of the model. However, be-cause we modeled the galaxies to very low surface bright-ness levels, and performed careful iterative modeling toachieve flat local background levels, such residual “edge”features were generally in the noise. In addition, spuriousdetections associated with the model edges would be re-moved by our point source selection criteria described be-low. Panels (b) through (d) of Figure 1 respectively showour combined isophotal models for the galaxies labeledin panel (a) plus two additional galaxies; the isophotalgalaxy models plus the residual background map; andthe final “residual image” after subtracting the galaxyand residual background models. For comparison, thestacked WFC3/IR F160W science image and residualimage following galaxy and background map subtractionare presented in Figure 2. Because of the smaller fieldof view, only NGC 4874 and the three other galaxies (la-beled) were modeled. Disky residuals are noticeable insome cases, but the subtracted images are generally quiteclean, revealing many faint sources.

3.2. Object Detection and GC Candidate Selection

Object detection and photometric measurements wereperformed on the final residual images using SExtrac-tor independently for each bandpass (i.e., in “single-image mode”). For the ACS photometry, we used theRMS weight images produced by Apsis as the SExtractorweight images (type MAP RMS). For the WFC3/IR F160Wphotometry, we used a variance map (type MAP VAR) con-structed from the inverse-variance image produced byMultiDrizzle, and including the photometric noise fromthe science data image itself. In order to flag bad pix-

NGC 4874 Globular Clusters 5

(a)

N

E

NGC 4874 (cD)

PGC 44644 (SA0)

PGC 44651 (S0)

PGC 44636 (SA0/a)

(b)

N

E

40′′ = 19.2 kpc

Figure 2. Stacked HST WFC3/IR F160W observation of NGC4874 and surrounding region. The field size is 2.′36× 2.′09; north is to theright and east is up. (a) The drizzled F160W science image; NGC4874 and three neighboring galaxies are labeled. (b) The same imageafter subtracting our isophotal models of the four large galaxies and a SExtractor-generated background map.

els, we made maps denoting blank image areas, pixelsclose to frame boundaries, and the circular detector de-fect visible in WFC3/IR images. The maps were refer-enced using FLAG IMAGE in SExtractor. We ran SExtrac-tor with a Gaussian detection filter to identify objectswith an area of at least four connected pixels with aflux level above two times the background rms in theACS F475W and F814W images. The slightly largervalue of DETECT MINAREA = 5 was used for the WFC3/IRF160W image since the subpixel resampling from theoriginal pixel scale to 0.′′1 pix−1 during the MultiDriz-zle run causes more noise correlation between neighbor-ing pixels. Separation of blended objects was performedusing the SExtractor parameter DEBLEND NTHRESH = 32and DEBLEND MINCONT = 0.005 and 0.007 for ACS/WFCand WFC3/IR images, respectively.The source catalogs extracted from the ACS/WFC

F475W and WFC3/IR F160W images were matchedagainst the ACS/WFC F814W catalog using the sourcepositions to remove spurious sources from the multi-band data. We estimate total I814 magnitudes for eachobject using the MAG AUTO values. For the color esti-mations, the aperture photometry was performed usingapertures with radii of 3 pixels (0.′′15 for ACS and 0.′′30for WFC3/IR data) as in Blakeslee et al. (2012). Aper-ture corrections were determined for a typical GC atthe Coma distance using PSF-convolved King models.The empirical PSFs for ACS/F475W, ACS/F814W, andWFC3/F160W bands were produced with the same driz-zle parameters, including interpolation kernel, pixfrac,and output scale (all of which have important effects formagnitudes measured within small apertures) as the sci-ence data for each band. Our final aperture correctionsfor the 3-pixel radius SExtractor apertures are −0.24,−0.26, and −0.28 mag for g475, I814, and H160, respec-tively, with uncertainties of 0.01 mag.In order to identify GC candidates, we used the F814W

photometric catalog because of its higher signal-to-noise

ratio (S/N) than the F475W data, and larger field ofview than the WFC3/IR image. Prior to classifying can-didates as GCs, we required the SExtractor parameterFLAGS < 4 for all three bands in order to exclude sourcestoo near the image edges (e.g., Puzia et al. 2014). Tolimit our analysis to sources detected with S/N > 5, werequire MAGERR ISO (the rms error on the magnitudeswithin the isophotal area) in F814W to be less than0.2 mag. Figure 3 shows our photometric selection cri-teria for probable GCs as a function of the total I814magnitudes, for which we adopt the values of MAG AUTO

measured with SExtractor (as an additional sanity check,we require the uncertainty in MAG AUTO to be less than1 mag). As demonstrated in the top left panel of Fig-ure 3, the uncertainties on the isophotal magnitudes aresmaller than 0.1 mag for the majority of the GC candi-dates brighter than the turnover of the GC luminosityfunction (GCLF), which is expected to occur at an ABmagnitude of I814 ≈ 26.9 ± 0.2 mag at the distance ofthe Coma cluster (Peng et al. 2009, 2011).The majority of GCs can be treated as point sources

in our HST images since the mean half-light radius oftypical GCs in early-type galaxies, rh ≈ 3 pc (Jordanet al. 2005, 2009; Masters et al. 2010), corresponds to∼ 0.′′006 at 100 Mpc. We therefore required candidateGCs to be compact. The SExtractor “stellarity index”values CLASS STAR for all detected objects in F814W areplotted against the total magnitudes in the bottom leftpanel of Figure 3. The objects with CLASS STAR > 0.5were classified as point-like sources, and thus possibleGCs in Coma. Since the CLASS STAR parameter is unre-liable for the fainter objects, we also adopted additionalcriteria, based on the measured FWHM and concentra-tion index C4−10, to select faint point-like sources. TheC4−10 concentration index was introduced by Peng et al.(2011) in order to select likely GC candidates in Coma; itis defined as the difference between magnitudes measuredin apertures with diameters of 4 pix and 10 pix. These

6 Cho et al.

Figure 3. Initial criteria for GC candidate selection as a function of I814 MAG AUTO (an estimate of the total magnitude) from SExtractor.Clockwise from top left panel: RMS error for the isophotal magnitude (scaling inversely with detection signal-to-noise), full width athalf-maximum, magnitude difference between 4- and 10-pixel diameter apertures, and the CLASS STAR stellarity index values are plottedagainst MAG AUTO. The gray hatched regions mark the parameter ranges over which objects are excluded under each criterion (see Section 3.2for details).

Figure 4. Spatial distributions of GC candidates in the magnitude range of 21.5 < I814 < 27.0 mag (red points) plotted on top of theACS/WFC F814W image. The left panel shows positions of the GC candidates selected only from the F814W photometry. The middlepanel shows GC candidates from the matched F814W and F475W photometric catalogs with colors 0.5 < g475−I814 < 1.6 mag and colorerrors < 0.2 mag. The right panel shows the positions of the GC candidates from the matched F814W and WFC3/F160W photometriccatalogs with colors −0.5 < I814−H160 < 1.5 mag, and color errors < 0.2 mag.

NGC 4874 Globular Clusters 7

additional selection criteria are graphically indicated inthe right panels of Figure 3, where it is clear that a largefraction of the detected sources follow tight loci arounda FWHM of 2 pix and a C4−10 value of 0.4 mag.We can thus summarize our initial (i.e., from the

ACS F814W band, prior to any color cuts) GC can-didate selection criteria as follows: MAGERR ISO< 0.2,1< FWHM< 4 pix (with 0.′′05 pix−1), and 0.0 < C4−10 <1.0 mag. We adopted a relatively broad cut in C4−10 inorder to include GC candidates that are more extendedthan typical GCs. However, using the above combina-tion of criteria ensures that we select robustly charac-terized compact sources as GC candidates. In Figure 4(left panel), we plot the locations in the ACS F814Wimage of the 6303 GC candidates selected solely fromthe F814W photometric data with magnitudes in therange 21.5 < I814 < 27.0 mag. These F814W GC candi-dates are widely distributed around the central cD galaxyNGC4874, with localized concentrations around severalof the surrounding cluster galaxies.In this work, we also analyze the color properties of

the GC candidates, and for this analysis we impose ad-ditional criteria to reject objects that are likely to becontaminants based on their color. In matching theF814W-selected candidates with the ACS F475W ob-ject catalog, we imposed a broad color cut of 0.5 <g475−I814 < 1.6 mag (e.g., Peng et al. 2011) for theGC candidates. We plot in the central panel of Fig-ure 4 the 4612 GC candidates from the left panel thathave colors within this range and color uncertainties lessthan 0.2 mag. In matching the F814W candidates tothe WFC3/F160W catalog, we restricted the colors to−0.5 < I814−H160 < 1.5 mag (e.g., Blakeslee et al. 2012);again requiring color uncertainties less than 0.2 mag, weplot the 1719 GC candidates within this I814−H160 colorinterval in the right panel of Figure 4. Note that thepaucity of matched F814W+F160W GC candidates nearthe galaxy NGC4873 (labeled in Figure 1) occurs be-cause this galaxy is off the edge of the WFC3/IR fieldof view (see Figure 2) and was not cleanly subtractedby isophotal modeling; thus, we did not obtain reliableF160W photometry for objects in its immediate vicinity.

3.3. Comparison with ACS Coma Cluster Survey

Photometry in the ACS g475 and I814 bands for GCsin the region around NGC4874 was previously publishedby Peng et al. (2011) using data from HST programGO-10861, ACSCCS. Our exposure time in F814W is7.4 times longer than that obtained by the ACSCCS, im-plying a S/N about 2.7 times greater, or a limiting mag-nitude more than 1 mag deeper in this band. For F475W,because we incorporated the ACSCCS exposures into ourstacked image, our exposure time is nearly a factor of twolonger (∼ 40% higher S/N) than for the ACSCCS dataalone (the images did not overlap completely becausethey were taken at different orientations). Since the ad-dition of the ACSCCS F814W data would have increasedour S/N by . 7%, we opted not to include those data inour stacked image in that bandpass. Peng et al. (2011)performed source photometry on the galaxy-subtractedACSCCS images using SExtractor with 3 pixel radiusapertures and then selected GC candidates based oncolor and source concentration; thus, the analysis wasquite similar to our own and can be used as a straight-

Figure 5. Comparison of 3-pixel radius aperture magnitudesin the ACS F814W passband from the photometry of Peng et al.(2011), denoted by IPeng, with those from our photometry. Thepoints are objects in common for the two data sets. The blacksolid line in the upper panel represents equality, and is not a fit.In the lower panel, the magnitude differences ∆I = IPeng − I814are plotted as a function of our I814 magnitude. The black solidline shows a zero magnitude difference, while the blue dashed lineindicates the median offset in ∆I of 0.017 mag.

forward check on our photometry.Figure 5 shows a comparison of I814 magnitudes from

Peng et al. (2011) with those of the present study; forconsistency, we compare the magnitudes without correc-tion for Galactic extinction. The data for this compari-son (unlike the case for g475) are fully independent. Thetop panel of the figure shows that the overall agreement isvery good over a range of 5 mag; the slope of the resid-uals over this magnitude range is consistent with zero.Peng et al. cut their GC selection at I814 = 26.5 mag, inpart because the photometric error in their C4−10 concen-tration parameter became too large to distinguish pointsources and background galaxies at about this magni-tude; as expected from the increased depth, our F814Wdata are able to distinguish point sources from extendedobjects to about 1 mag fainter (compare their Figure 2with our Figure 3).The lower panel of Figure 5 shows the residuals ∆I =

IPeng − I814 for 2673 point sources (selected based on ourmeasurements) in common between the two data setsdown to I814 = 26.5 mag (again, from our measurement).The mean offset in I814 is 0.014±0.002 mag, with a scat-

8 Cho et al.

Figure 6. Comparison between our g475−I814 color values and those from the photometry of Peng et al. (2011), denoted by(g475−I814)Peng. Black points have color errors in the Peng et al. photometry smaller than 0.2 mag. The objects with larger coloruncertainties are marked by gray points. The black solid lines in the upper panels represent the one-to-one relation. The lower panels showthe differences in the colors between these two data sets. In each lower panel, the zero and median difference values are marked by theblack solid line and the blue dashed line, respectively. The red solid lines show the robust linear relations given in the text.

ter of 0.105 mag, and the median offset is 0.017 mag,in the sense that the ACSCCS magnitudes are slightlyfainter. If we limit the range to 21.5 < I814 < 25.5 mag,then the number of sources is reduced by approximatelyhalf to 1308, with both a mean and median offset of0.022 mag, and a scatter of 0.063 mag. Peng et al. (2011)calibrated their photometry using the AB zero pointsfrom Sirianni et al. (2005); however, adopting the cali-bration from the online ACS zero point calculator for theappropriate date of the observations would decrease thesize of the offset by only 0.002 mag. Given the uncer-tainty in the time-dependence of the zero point (Bohlin2012), the lack of CTE correction for the ACSCCS data,the difference in the drizzle parameter settings (e.g., lin-ear versus lanczos3 interpolation kernels), the possibilityof small focus variations (e.g., Jee et al. 2007), and themuch greater depth of our F814W observations (whichcould result in subtle differences in the SExtractor pho-tometry), we consider the systematic offset of . 0.02 magto be reasonable. The scatter in the residuals (domi-nated by the much shallower ACSCCS measurements)increases as expected at fainter magnitudes, but there isno evidence for a systematic trend in the residuals withmagnitude.Figure 6 compares the g475−I814 colors for point

sources in common between Peng et al. (2011) and thepresent study over a range in I814 from 21.5 to 26.5 mag.The left and right panels show, respectively, the com-parison as a function of our and the ACSCCS colormeasurements. In both panels, the black points repre-sent sources in Peng et al. (2011) with estimated colorerrors < 0.2 mag, while the gray points show sourceswith color errors larger than 0.2 mag. Considering allthe points, black and gray, over the plotted color rangeof 0.45 < g475−I814 < 1.65 mag, the median offset is0.026 mag, the rms scatter is 0.13 mag, and the bi-weight scatter (more robust against outliers) is 0.12 mag.Considering just the black points, the median offset is0.020 mag, the rms scatter is 0.11 mag, and the biweightscatter is 0.10 mag. The sense of the offset is that theACSCCS g475−I814 colors are slightly redder than ours;if we were to recalibrate the Peng et al. photometry us-ing the online ACS Zeropoints Calculator, the ACSCCScolors would become bluer by 0.021 mag, reducing themedian color offset for the black points to 0.001 mag.However, because of observational error, the observedoffset also has a dependence on color.The lower panels of Figure 6 show the color differences

∆(g475−I814) (defined as y-axis color minus x-axis color)plotted as a function of both our colors and the ACSCCS

NGC 4874 Globular Clusters 9

colors, which we label (g475−I814)Peng. The solid redlines show robust linear regressions for the black pointsin these panels; the slopes of the ∆(g475−I814) regressionlines are −0.068± 0.012 and −0.214± 0.010 when fittedversus our g475−I814 colors and versus (g475−I814)Peng,respectively. Thus, the slope is more than a factor ofthree steeper when fitted as a function of the ACSCCScolors. This is understandable in light of the larger mea-surement errors for those colors. Since the vast major-ity of these objects are GCs, which intrinsically define afairly narrow color range 0.7 . g475−I814 . 1.4 mag, thescattering of the colors outside this color range primar-ily results from photometric errors, which are larger forthe shallower ACSCCS measurements; thus, this error-induced slope is larger when plotted as a function of theACSCCS colors. We find that we can reproduce theslopes and scatters in Figure 6 if we assume Gaussianerrors with σ = 0.055 mag for our color measurementsand σ = 0.107 mag the ACSCCS colors. For compari-son, the median estimated color errors in the two catalogsare 0.063 mag and 0.098 mag, respectively. This suggeststhat our quoted errors may be slightly overestimated andthe ACSCCS color errors slightly underestimated, butonly by about 10% in each case.We conclude that our measurements agree well with

the ACSCCS photometry from Peng et al. (2011). Themuch greater exposure time of our I814 imaging allowsus to reach about 1 mag fainter in this bandpass, whileour g475−I814 color errors for GC candidates are approx-imately a factor of two smaller than for the ACSCCSdata. Systematic offsets in photometry are . 0.02 mag.In addition, our program adds deep H160 photometryover the area of WFC3/IR field, which was not availablefor the earlier study.

4. DISCUSSION

4.1. Color-Magnitude Diagrams and “Tilts”

As discussed in the Introduction, GC systems of mas-sive galaxies generally follow bimodal distributions in op-tical color. However, the peaks in the color distributioncan vary with the magnitude range of the GCs consid-ered. For instance, Ostrov et al. (1998) and Dirsch et al.(2003) found that for GCs more than 2 mag brighterthan the turnover of the GCLF in the galaxy NGC1399,the blue and red peaks merged together into a singlebroad distribution. More generally, the mean color ofthe blue GCs tends to get redder at brighter magnitudes(Harris et al. 2006; Mieske et al. 2006, 2010; Straderet al. 2006; Harris 2009), possibly indicating an increas-ing mean metallicity with GC luminosity. This effect,known alternately as the GC color-magnitude relation,mass-metallicity relation, or informally as “the blue tilt”(a “tilt” of the blue peak towards a redder mean colorat bright magnitudes) is most generally believed to be aconsequence of self-enrichment within the most massiveGCs (e.g., Bailin & Harris 2009).Figure 7 displays the color-magnitude diagrams

(CMDs) for GC candidates in NGC4874 in both the op-tical g475−I814 (left panel) and optical-NIR I814−H160

(right panel) colors. We have marked in these panelsthe objects, included in our GC selection, that are spec-troscopically confirmed (red squares) or possible (bluetriangles) ultra-compact dwarfs (UCDs) from the study

of Chiboucas et al. (2011). In this case, UCDs are de-fined simply as compact stellar systems with colors sim-ilar to GCs and absolute R magnitude MR < − 11 mag.Because of the larger ACS/WFC field, there are six ofthese objects in the optical CMD of the left panel, butfour in the right panel (in both cases, two of the UCDs areuncertain Coma members based on their spectra). Wealso indicate objects (green circles) that were selectedby Madrid et al. (2010) based on ACSCCS imaging ascandidate (lacking spectroscopic confirmation) UCDs or“dwarf-globular transition objects,” defined as objectshaving GC-like colors and half-light radii in the range of10 to 100 pc (if located at the distance of the Coma clus-ter). Although there are more objects in the left panel,and the ridge-line of the blue GC component is also muchmore distinct in the g475−I814 color, overall the CMDsappear fairly similar over a range of 5 mag in luminosity,with an overall tilt towards redder colors at the brightestmagnitudes where objects tend to be classified as UCDs.In order to quantify the degree of “blue tilt” in

NGC4874, we binned the GC candidates by magnitudeand applied the Gaussian Mixture Modeling (GMM)code of Muratov & Gnedin (2010) to each bin. Figure 8shows CMDs similar to the previous figure, but now us-ing I814 for the magnitude in both cases, and showingthe locations of the color peaks from the bimodal GMMdecompositions for eighteen bins in magnitude down toI814 ≈ 26.5 mag. Because the GMM algorithm can besensitive to objects that are scattered into the tails of thedistribution by observational errors (which increase atfainter magnitudes), we have restricted these magnitude-grouped samples to the color ranges 0.6 < g475−I814 <1.5 mag and −0.1 < I814−H160 < 1.1 mag, indicated bythe dotted lines in Figure 8. For the I814 versus g475−I814CMD (left panel), each bin has 240 GCs, while for theI814−H160 CMD (right panel), each bin has 90 GCs; theexceptions are the brightest two bins in each panel, whichhave only half the number of GCs as the other bins. Inthe left panel of Figure 8, there is clear evidence for a“blue tilt,” as well as some suggestion of a “red tilt” forthe peak positions at magnitudes I814 < 25 mag, corre-sponding to absolute M814 < −10 mag. Linear fits to thered and blue peak positions for bins brighter than thismagnitude are shown by the solid black lines, defined bythe following relations:

(g475−I814)blue=(0.88± 0.01) − (0.082± 0.020)

×(I814 − 24.5), M814 < −10.0 ; (1)

(g475−I814)red=(1.14± 0.01) − (0.031± 0.023)

×(I814 − 24.5), M814 < −10.0 . (2)

The error bars here reflect the statistical uncertainties inthe parameters from the linear fits.The slope γ814,blue ≡ d(g475−I814)/dI814 = −0.082

of the blue tilt is highly significant (4-sigma) and isamong the steepest observed to date. For compar-ison, Mieske et al. (2006, 2010) found γ850,blue ≡d(g475−z850)/dz850 = −0.042 ± 0.015 for M87 in Virgoand γ850,blue = −0.088± 0.025 for NGC1399 in Fornax,the central giant ellipticals in each cluster. Using theobserved relationship for GCs in NGC1399, g475−I814 =0.13 + 0.75(g475−z850) from Blakeslee et al. (2012), thiswould imply γ814,blue = −0.032 and −0.065 for M87 andNGC1399, respectively. Further, NGC4472 (M49), the

10 Cho et al.

Figure 7. The optical color-magnitude diagram for GC candidates in the NGC4874 field from ACS/WFC F475W and F814W imagingdata (left) and the optical-NIR color-magnitude diagram for GC candidates from ACS/WFC F814W and WFC3/IR F160W imaging data(right). The final GC samples from Figure 4 are plotted in black; the gray points were excluded from further analysis. Error bars (nearthe right edge of each panel) represent the mean errors of the magnitudes and colors in a magnitude bin. Red squares show ultra-compactdwarf galaxies (UCDs, classified purely on the basis of luminosity and color) spectroscopically confirmed as members of the Coma cluster,while blue triangles indicate likely UCDs with uncertain redshifts (Chiboucas et al. 2011). Green circles mark photometrically classifiedUCDs or “dwarf-globular transition objects” from Madrid et al. (2010). These objects all have photometric properties consistent withbeing extensions of the GC population, and we do not attempt to exclude them from our analysis.

brightest galaxy in Virgo, has no significant blue tiltat all; thus, the color-magnitude trend in NGC4874 isexceptionally steep compared to the Virgo and Fornaxclusters. This may be related to an abundance of UCDsin the dense core of the Coma cluster. It is clear thatthe tilt becomes greater for objects at I814 < 23.5 mag(M814 < −11.5 mag), where the sample may be domi-nated by UCDs, which tend to have colors intermediatebetween the blue and red peaks of the optical GC colordistribution (e.g., Liu et al. 2015). It is likely that UCDsrepresent a mix of stripped galactic nuclei and luminousGCs; as already indicated in Figure 7, we have not at-tempted to exclude UCDs from our sample if they satisfyour selection criteria.The derived color-magnitude slope becomes markedly

less steep when the fit is extended to fainter GCs. Thedashed lines in Figure 8 indicate the following linear fitsto the peaks with I814 < 26 mag (M814 < −9 mag):

(g475−I814)blue=(0.88± 0.01) − (0.027± 0.010)

×(I814 − 24.5), M814 < −9.0 ; (3)

(g475−I814)red=(1.15± 0.01) + (0.012± 0.015)

×(I814 − 24.5), M814 < −9.0 . (4)

Thus, when the fit is extended by one magnitude, theslope of the color-magnitude trend for blue peak positionsis reduced by a factor of three. The shallower slope overthe wider magnitude range reflects the nonlinearity ofthe color-magnitude tilt (e.g., Harris 2009; Mieske et al.2010), which may result from a minimum mass threshold

for self-enrichment. For the red GC peak, the fitted slopeover this broader magnitude range is essentially zero.We can estimate the scaling of metallicity Z with the

GC luminosity L in the I814 bandpass using the em-pirical broken-linear calibration from Peng et al. (2006)for the metallicity as a function of g475−z850 color.Since the “tilt” occurs for the blue GC population, weuse the linear relation appropriate for the blue GCs:[Fe/H] = −6.21 − (5.14 ± 0.67)(g475−z850). This is thesame relation used by Mieske et al. (2010) for derivingthe mass-metallicity scaling from their blue tilt measure-ments in the ACS Virgo and Fornax Cluster Survey data(Cote et al. 2004; Jordan et al. 2007). Coupled withthe above relation between g475−I814 and g475−z850,and our measurement of γ814,blue = −0.082 ± 0.020 forM814 < −10 mag, we find Z ∝ L1.4±0.4 at these high-est luminosities, or if we assume a constant mass-to-light ratio for blue-peak GCs as in Mieske et al., thenZ ∝ M1.4±0.4

GC for the scaling with GC mass. Of course,if we use the slope γ814,blue = −0.027 ± 0.010 from thelinear fit extending to M814 = −9 mag, then the meanmass-metallicity scaling over this magnitude range be-comes Z ∝ M0.5±0.2

GC , again reflecting the nonlinearity ofthe relation.For the I814 versus I814−H160 CMD (Figure 8, right

panel), we find no significant evidence for a “tilt” in thecolors of either the red or blue peaks from the GMMbimodal decompositions within the magnitude bins. Thismay be because of the poorer statistics and/or weakerseparation of blue and red GCs for this optical-NIR color.

NGC 4874 Globular Clusters 11

Figure 8. I814 vs. g475−I814 color-magnitude diagram (left) and I814 vs. I814−H160 color-magnitude diagram (right). The blackpoints in Figure 7 are plotted in gray in these diagrams for clarity. We subdivided each CMD into eighteen magnitude bins, each with afixed number of data points, except the brightest two bins in each panel, for which the number is half that of the other bins. In the leftpanel, the positions of the first and second peaks (mean positions) in the GMM double Gaussian model for each magnitude bin are plottedwith blue open diamonds and red open squares, respectively. The error bars represent the uncertainties on each peak calculated from thenon-parametric bootstrap resampling by the GMM algorithm. In the right panel, the green circles with error bars indicate the positionsof the average peaks for each magnitude bin. We also plot the positions of the double peaks from the GMM fits, which are marked byblue open diamonds and red open squares, along with the error bars from the GMM bootstrap resampling. The thick solid and dashedlines indicate the linear fits to the peak positions when the faintest magnitudes used in the each fit are M814 = −10.0 mag and −9.0 mag,respectively. For the fits, we adopted the blue and red color limits marked by the vertical dotted lines.

Notably, however, we do find a significant trend for theoverall mean GC color (based on the unimodal GMM fit)to become redder for the brighter magnitude bins. Thesolid line in this panel is a fit to the I814−H160 unimodalpeak positions for bins withM814 < −10 mag; the dashedline again extends the fit one magnitude fainter than thisand is significantly less steep. The fits are given by thefollowing relations:

(I814−H160)mean=(0.47± 0.01) − (0.093± 0.013)

×(I814 − 24.5), M814 < −10.0 ; (5)

(I814−H160)mean=(0.50± 0.01) − (0.034± 0.010)

×(I814 − 24.5), M814 < −9.0 . (6)

The slope of this “mean tilt” in I814−H160 for M814 <−10, I814 < −25 mag, is highly significant. The colorsequence appears nearly vertical at magnitudes fainterthan this, although the slope of the fit for M814 <−9 mag (dashed line) remains significant because of thebrightest bins with their increasingly steep slope. Al-though the relation between I814−H160 and metallic-ity has not been empirically calibrated for extragalac-tic GC systems, we can check for consistency by usingthe linear version of the relation between I814−H160 and

g475−I814 derived in Sec. 4.3 below, which has a sloped(I814−H160)/d(g475−I814) = 1.13 ± 0.04. Combiningthis with the same set of relations between g475−I814,g475−z850, and [Fe/H] as above (although the adopted[Fe/H] transformation is only strictly applicable for blueGCs), we can derive the mass-metallicity scaling from thefitted slopes d(I814−H160)/dI814 in Eqs. (5) and (6). ForM814 < −10 mag, the result is again Z ∝ M1.4

GC, and theexponent again drops to ∼ 0.5 if we use the I814−H160

fit extending to M814 = −9 mag.The equality in the exponents of the mass-metallicity

relations derived from g475−I814 and I814−H160 mayseem strange, given that in former case it is based on thetrend in the blue GC component with magnitude, whilefor the latter it is based on the overall mean I814−H160

trend with magnitude. In fact, it is somewhat fortu-itous. The ratio of the slope for the mean I814−H160

in Eq. (5) to the slope for the blue peak in Eq. (1) is1.13± 0.28; the corresponding ratio for Eqs. (6) and (3)is 1.26± 0.42. Both of these are in statistical agreementwith the color-color slope found below in Sec. 4.3. Thisagreement can be understood, at least in part, from thefact that at progressively brighter magnitudes, the pro-

12 Cho et al.

0.12

0.08

0.04

0.00

d(g−I

)

dI

24.5 25.0 25.5 26.0 26.5I814, cut_faintest [ABmag]

0.12

0.08

0.04

0.00

d(I−H

)

dI

10.5 10.0 9.5 9.0 8.5M814 [ABmag]

Figure 9. Fitted slopes as a function of the faintest magnitudecut in the ACS F814W passband. The red squares and blue di-amonds in the top panel are the slope values of the linear fits tothe red and blue peak positions, respectively, in the g475−I814 ver-sus I814 CMD shown in Figure 8 for varying limiting magnitudesof the fit. The green points in the bottom panel are the slopesof the linear fits to the overall mean positions in the I814−H160

versus I814 CMD in Figure 8 for varying limiting magnitudes ofthe fit. The vertical solid and dashed lines indicate the limitingmagnitude values for the corresponding linear fits shown in Fig-ure 8. The trends seen here towards shallower slopes with fainterlimiting magnitudes indicate that the relations between mean colorand magnitude are nonlinear, with the color peak positions in theCMDs becoming more vertical at fainter magnitudes; put anotherway, the “tilts” are only significant for the brightest GCs.

portion of red-peak to blue-peak GCs increases in theg475−I814 histogram, as shown in Sec. 4.2 below. Thus,the overall mean slope of g475−I814 versus I814 will besteeper than the average of the red and blue slopes. Forcompleteness, we also fitted the overall mean g475−I814color-magnitude relations, finding:

(g475−I814)mean=(1.02± 0.01) − (0.060± 0.014)

×(I814 − 24.5), M814 < −10.0 ; (7)

(g475−I814)mean=(1.03± 0.01) − (0.027± 0.008)

×(I814 − 24.5), M814 < −9.0 . (8)

In both cases, the slope is steeper than the average of thered and blue slopes derived for the equivalent magnitudelimits. In fact, the slope we find for the mean trend inEq. (8) is the same as that for the blue tilt in Eq. (3).However, the conversion of these mean trends to a mass-metallicity scaling relation is less straightforward becausethere is a change in the slope of the color-metallicityrelation at intermediate colors (e.g., Peng et al. 2006;Usher et al. 2012).Figure 9 explores in more detail the dependence of the

slope of the color-magnitude tilts as a function of thefaint limit of the linear fits. The slope of the blue peakin g475−I814 remains significant regardless of the magni-tude limit, while the slope for the red peak appears signif-

icant at the 2σ level only when the brightest two or threebins are considered, those with M814 < −10.3 mag. ForI814−H160 (lower panel), although the separation intoblue and red peaks is weak (as quantified in the follow-ing section), for bins with M814 < −8.8 mag, the slope ofthe overall trend towards a redder mean color (and thusmetallicity) at brighter magnitudes is quite significant.However, the magnitude of the slope decreases contin-uously from M814 ≈ −10 to M814 ≈ −8.5 mag, againillustrating the nonlinearity of the trend.

4.2. Color Distributions

As discussed in the previous section, the NGC4874 GCcandidates exhibit a distinct color-magnitude relation, atleast at I814 < 25 mag. Consequently, their color distri-butions should vary as a function of luminosity. Figure 10plots the g475−I814 optical color histograms for the GCcandidates in four different bins in I814 magnitude; thedistributions differ markedly from each other. In thebrightest bin, consisting of objects at least 4 mag brighterthan the expected turnover of the GCLF, the distribu-tion is relatively red and broad, with no evidence for bi-modality. The spectroscopic sample of UCDs from Chi-boucas et al. (2011) is weighted towards the blue side ofthis distribution, but the sample is small and incomplete(see Figure 7), and selection effects could play a role. Inthe second bin, 23.0 < I814 < 24.0 mag, there is clearbimodality in g475−I814, with the red peak being domi-nant. For 24.0 < I814 < 25.0 mag, the blue peak becomesdominant, and this is true to an even greater extent forthe faintest magnitude bin of 25.0 < I814 < 26.0 mag.Figure 11 shows the corresponding histograms of

I814−H160 color using the same magnitude bins as in Fig-ure 10. The samples are smaller because of the smallerfield of WFC3/IR, but again we find that the color distri-bution appears broad, unimodal, and red for the bright-est magnitude bin. Although any bimodality is muchless evident than in g475−I814, the I814−H160 histogramfor the 23.0 < I814 < 24.0 mag range is skewed towardsthe red, while the histograms for the faintest two plot-ted magnitude ranges become progressively more skewedtowards the blue. This is qualitatively similar to whatis observed for g475−I814, and it is consistent with thestriking “mean tilt” in the I814−H160 versus I814 color-magnitude relation (Figure 8), for which the colors be-come bluer in I814−H160 at fainter magnitudes.In order to quantify the visual impressions given by

Figures 10 and 11, we ran the GMM code on theg475−I814 and I814−H160 color distributions of the GCcandidates in the various magnitude ranges shown inthose figures. Table 2 summarizes the results of theseGMM analysis runs, as well as the results for the broadermagnitude range of 23.0 < I814 < 25.0 mag, for whichthe g475−I814 and I814−H160 histograms are displayedin Figure 12. The bimodality in g475−I814 is significantfor all the magnitude ranges explored in Table 2 exceptfor the brightest; all the other bins have p(χ2) < 0.01,indicating less than 1% probability of the color data be-ing drawn from a single Gaussian model, rather thanthe best-fit double Gaussian model with the tabulatedmeans µ1, µ2 and dispersions σ1, σ2 and with the frac-tion of objects in the second (red) Gaussian given by f2.The evidence for bimodality is stronger if the tabulatedD, the separation between the Gaussians in units of the

NGC 4874 Globular Clusters 13

0

20

40 21.5<I<23.0

0

20

40 23.0<I<24.0

0.5 1.0 1.50

50

10024.0<I<25.0

0.5 1.0 1.50

100

20025.0<I<26.0

g475−I814 [ABmag]

Freq

uenc

y

Figure 10. Histograms of g475−I814 colors for GC candidates in the left panels of Figures 7 and 8 over different I-band magnitude ranges.The smooth Gaussian kernel density estimates are overplotted by thick solid curves. The hatched histogram is for UCDs (both confirmedcluster members and uncertain ones) from Chiboucas et al. (2011).

0

10

20

30 21.5<I<23.0

0

10

20

30 23.0<I<24.0

0.0 0.5 1.00

50

100 24.0<I<25.0

0.0 0.5 1.00

50

100 25.0<I<26.0

I814−H160 [ABmag]

Freq

uenc

y

Figure 11. Same as Figure 10 but for I814−H160 colors for GC candidates in the right panels of Figures 7 and 8.

quadrature sum of their dispersions, is significantly > 2,and if the kurtosis of the distribution kurt < 0 (see Mu-ratov & Gnedin 2010 and Blakeslee et al. 2012). The op-tical bimodality is especially pronounced, and the doubleGaussian model parameters best constrained, within the23.0 < I814 < 25.0 mag range.For the I814−H160 color index, the bimodality is only

significant at the > 2σ level, p(χ2) < 0.05, for the

23.0 < I814 < 25.0 mag range. Interestingly, how-ever, for this magnitude range, the GMM code givesf2 = 0.624± 0.055 for g475−I814, but f2 = 0.330± 0.144for I814−H160. Thus, although the bimodality is signif-icant in this magnitude range for both g475−I814 andI814−H160, the preferred ratios of red/blue GCs differ atthe ∼ 2σ level. This is similar to the result for NGC1399,the cD galaxy in the Fornax cluster, for which Blakeslee

14 Cho et al.

0.5 1.0 1.5g475−I814 [ABmag]

0

40

80

120

Freq

uenc

y Ntotal = 939

0.0 0.5 1.0I814−H160 [ABmag]

Ntotal = 393

Figure 12. Histograms of g475−I814 and I814−H160 colors for GC candidates within the magnitude range of 23.0 < I814 < 25.0 mag.The black thick solid curve is the nonparametric density estimate constructed with a Gaussian kernel. We also plot the GMM doubleGaussian model components for the heteroscedastic case with blue and red solid curves. The corresponding GMM analysis results areprovided in Table 2.

0.5 1.0 1.5g475−I814 [ABmag]

0

40

80

120

Freq

uenc

y Ntotal = 392

0.0 0.5 1.0I814−H160 [ABmag]

Ntotal = 392

Figure 13. Same as Figure 12 but for the cross-matched subsample of GC candidates with both ACS/WFC F475W and WFC3/IRF160W data. The GMM analysis results are in Table 3.

et al. (2012) found significantly different bimodalities ing475−I814 and I814−H160, resulting from the nonlinearrelation between these two color indices. However, itshould be noted that although the magnitude range isthe same, the g475−I814 sample has 939 objects whilethe I814−H160 sample has only 393 objects because theWFC3/IR field of view is smaller; it is not clear if the dif-ference in color bimodalities is significant or not becausethe samples are different.Figure 13 shows the histograms for the cross-matched

subsample of 392 GC candidates in the 23.0 < I814 <25.0 mag range having both g475−I814 and I814−H160

colors. (There was one object in the sample of GC can-didates with I814−H160 colors that was not included inthe sample with g475−I814 colors.) The optical g475−I814color is clearly bimodal, while the separation remainsless clear for I814−H160. Table 3 presents the GMManalysis results for the g475−I814 and I814−H160 dis-tributions of this homogeneous cross-matched sample.We include both the homoscedastic (common dispersion,σ1 = σ2) and heteroscedastic (σ1 6= σ2) cases. For theheteroscedastic case, the preferred bimodal decomposi-tions again differ significantly, with f2 = 0.604 ± 0.083

and f2 = 0.324 ± 0.128 for g475−I814 and I814−H160,respectively. On the other hand, for the homoscedas-tic case, the GMM code finds f2 = 0.493 ± 0.031 andf2 = 0.436± 0.054 for g475−I814 and I814−H160, respec-tively. Thus, if the color dispersion for the blue and redGC components are forced to be the same, then the bi-modal decompositions for g475−I814 and I814−H160 areconsistent. However, the heteroscedastic GMM resultsimply that the dispersions differ significantly, at least forthe purely optical g475−I814 color, with the blue peakbeing significantly narrower; the same result has beenfound for other massive galaxies (e.g., Peng et al. 2006,2009; Harris et al. 2016).The heteroscedastic GMM results for I814−H160 in Ta-

bles 2 and 3 indicate that the color dispersion is slightlylarger for the blue component than for the red compo-nent, the opposite of what we find for g475−I814. Explor-ing this issue in more detail, we found that the disper-sion of the blue component in I814−H160, as well as theblue:red ratio, was sensitive to the presence of a smallnumber of GC candidates with the bluest I814−H160 col-ors. Table 4 reports the results for heteroscedastic GMMtests when the two and four bluest GCs in I814−H160 are

NGC 4874 Globular Clusters 15

removed from the sample. For instance, when the bluelimit is changed by +0.15 mag in I814−H160, reducingthe sample size from 392 to 388, the blue component ofthe GMM decomposition becomes significantly narrowerand the preferred red fraction goes from f2 = 0.32± 0.13to f2 = 0.54 ± 0.13, which is consistent with the f2 =0.61 ± 0.08 found for g475−I814. Thus, unlike the casefor NGC1399 (Blakeslee et al. 2012), the GMM decom-positions of the matched sample are consistent for theoptical g475−I814 and optical-IR I814−H160 colors, afterremoving a few of the bluest objects. However, we em-phasize that the GMM decomposition is not very robustfor I814−H160, mainly because the separation D of theblue and red components is not significantly greater thantwo, and thus any bimodality is difficult to quantify.

4.3. The Color-Color Relation

We now explore the relation between the sets of colormeasurements presented in the previous sections. Asdiscussed by Blakeslee et al. (2012), optical and mixedoptical-NIR color indices probe different spectral regionsand therefore different properties of unresolved stellarsystems. The g475−I814 color is sensitive to the main se-quence turnoff (which depends on the turnoff mass, andthus on age), the horizontal branch morphology (whichbehaves nonlinearly with metallicity and also depends onage; Lee et al. 1994; Dotter et al. 2010), and the tem-perature of the red giant branch. The I814−H160 coloris primarily sensitive to the temperature of the red giantbranch, which mainly depends on metallicity (e.g., Berg-busch & VandenBerg 2001; Dotter et al. 2007). Assum-ing similarly old ages for all the GCs, the form of therelation between different color indices reveals whetherthe colors behave differently as a function of metallicity,and thus can provide information on the color-metallicityrelations.Figure 14 shows I814−H160 as a function of g475−I814

for the matched sample of GC candidates in NGC4874.We plot only objects in the 23.0 < I814 < 25.0 magrange, where the color errors are small and the opticalbimodality is most pronounced. The figure shows thebest-fit bisector line, i.e., the linear relation that mini-mizes the orthogonal squared deviations, with 3-σ clip-ping; the clipped points are plotted as open circles. Theinset box in Figure 14 shows that a normal distributionwith dispersion σ ≈ 0.06 mag provides a good represen-tation of the orthogonal color residuals after clipping. Inorder to study deviations from a linear color-color re-lation, we grouped the data into twelve bins along thebisector line; the black squares in the upper panel ofFigure 14 show the modal locations within each bin, andthe lower panel shows the orthogonal deviations of thesebins from the linear relation. Eight of the twelve pointsdeviate significantly from the linear relation, followingan inflected, or “wavy,” locus at least qualitatively sim-ilar to the results found in other studies of the relationsbetween optical and optical-NIR GC colors using high-quality photometric data sets (e.g., Blakeslee et al. 2012;Chies-Santos et al. 2012; Cantiello et al. 2014).The binned modal values of the relation between

I814−H160 and g475−I814 are again shown in Figure 15,along with several different polynomial fits. Similar toour previous work (Blakeslee et al. 2012), the fits are ro-bust orthogonal regressions, weighted by the uncertain-

Figure 14. Optical-NIR I814−H160 vs. purely optical g475−I814color. The upper panel shows the color-color plane for individualGCs (solid and open gray points) and the modal (most probable)values within twelve bins (dark points with error bars) along withthe bisector line (solid black line). Outlier rejection was done by aniterative 3σ-clipping procedure based on minimizing the orthogonaldistances of the data points from the linear bisector fit assumingnormal distributions, as shown in the inset box (see Section 4.3for details). The filled gray circles are the final data points afteroutlier rejection, while the open circles show rejected outliers. Theblue dashed curve in the inset box is a normal distribution withσinitial = 0.071 for the light-blue shaded histogram before clipping.The black solid curve in the inset box is a normal distribution withσfinal = 0.056 mag for the black hatched histogram after clipping.The bin width along the bisector line was chosen to be 3σfinal, andthe bin spacing is half of the bin width. The lower panel shows theorthogonal deviations (black squares with error bars) of the twelvebinned data points with respect to the best-fit line in the upperpanel. The plotted points indicate the modal values within eachbin.

ties on the individual binned values; we also show the1-σ uncertainty regions around the fits. The equationsfor the plotted linear, cubic, and quartic fits are, respec-tively:

I814−H160=(−0.68± 0.04) + (1.13± 0.04)x , (9)

I814−H160=(−7.28± 1.53) + (20.60± 4.45)x

+(−18.91± 4.28)x2 + (6.05± 1.36)x3 , (10)

I814−H160=(−37.11± 6.17) + (136.11± 23.59)x

+(−185.27± 33.60)x2 + (111.68± 21.12)x3

+(−24.95± 4.94)x4 , (11)

where x ≡ g475−I814. Both the cubic and quartic poly-

16 Cho et al.

Figure 15. I814−H160 vs. g475−I814 color-color relation withweighted fits that minimize the squared orthogonal distances of thebinned data points (black squares, same as plotted in the upperpanel of Figure 14). The coefficients of each fit are given in thetext. The 1-σ ranges of the various fits are indicated with shadedregions. The black solid curve shows the relation from Blakesleeet al. (2012) and the black dashed curve indicates the relationshifted by −0.1 mag in I814−H160 (see text).

nomials provide statistically acceptable (within ∼ 1σ) de-scriptions of the data over the applicable domain 0.8 .g475−I814 . 1.3, while the linear fit is rejected with morethan 99.9% probability. Thus, the relation is nonlinearto a high degree of significance.The quartic fit derived for globular clusters in the For-

nax cD galaxy NGC1399 (Blakeslee et al. 2012) is alsoplotted in Figure 15. Unlike in the present analysis, the3-pixel aperture magnitudes from that study were notaperture corrected. This is a small effect for g475−I814because both g475 and I814 are measured on ACS datawith the same pixel scale and similar PSFs; thus, thedifferential aperture correction between the two bands issmall. However, it is a much larger effect for I814−H160

because the stacked WFC3/IR images have twice thepixel scale of the stacked ACS images, and GCs are sig-nificantly resolved at the 20 Mpc distance of the Fornaxcluster. Assuming King model profiles with the range ofhalf-light radii for GCs in the ACS Fornax Cluster Sur-vey (Masters et al. 2010), we find that the correction inI814−H160 would be in the range of ∼ 0.05 to ∼ 0.1 mag.Figure 15 shows that shifting the uncorrected 3-pix aper-ture color relation for NGC1399 by 0.1 mag provides anapproximate (though not statistically acceptable) matchto the NGC4874 relation. This remaining disagreementmay result from still larger differential aperture effectsat the blue end, where GCs tend to have larger sizes(Jordan et al. 2005; Masters et al. 2010), and/or intrinsicdifferences in the color-color relations and the underly-ing color-metallicity relations. Usher et al. (2015) foundthat there are significant differences in the g−i color-metallicity relations for different galaxies. We plan to

address this issue fully in a future paper presenting theHST optical-IR colors of GCs in a larger sample of Virgoand Fornax cluster galaxies, including detailed modelingof the differential aperture effects at these more nearbydistances. For now, we conclude that the relation be-tween g475−I814 and I814−H160 for GCs in NGC4874appears to have less extreme curvature than our previ-ously published relation for NGC 1399, but the deviationfrom a purely linear relation remains highly significant.

4.4. Radial Distributions

The spatial distributions of GCs around galaxies andwithin galaxy clusters can provide information on thebuildup of galaxy halos and cluster dynamical histories(e.g., Moore et al. 2006; Mackey et al. 2010; Lee et al.2010; Keller et al. 2012). Evidence for sizable popula-tions of IGCs, objects bound to the overall cluster po-tential rather than any individual galaxy, has been foundin several massive galaxy clusters, including Virgo (Leeet al. 2010; Durrell et al. 2014), Coma (Peng et al. 2011),Abell 1185 (West et al. 2011), and Abell 1689 (Alamo-Martınez et al. 2013). Numerical studies (Bekki & Ya-hagi 2006; Smith et al. 2013, 2015; Mistani et al. 2016)find that dwarf galaxies can lose substantial fractionsof their GC systems to the larger cluster environment,but they come to varied conclusions regarding whetherthe bulk of the IGC population results from stripping ofdwarfs or the outskirts of more massive galaxies. Penget al. (2011) found an extensive population of IGCs in thecenter of the Coma cluster. Because these IGCs showed asignificant tail of red GCs comprising roughly 20% of thepopulation, the authors concluded that a sizable fractionof the IGCs originated in massive galaxies, rather thanfrom disrupted dwarfs.In order to quantify the spatial distribution of the

GCs, and differences between the red and blue sub-components, we analyzed the projected surface num-ber density profiles of the ACS GC candidates in the23.0 < I814 < 26.0 mag range as a function of galac-tocentric radius R. (Note that we have not integratedthe observed counts over an assumed GCLF, in contrastto Peng et al. 2011.) From the center of NGC4874 toR = 150′′, the GCs were binned within fixed radial annuliof 10′′ width. The number of GCs within each annuluswas normalized by the effective area of the annulus to getthe number densities, and these are plotted against R inFigure 16, along with their Poisson-based uncertainties.We fitted the number densities with the commonly usedSersic (Sersic 1963) profile:

NGC(R) = Ne exp

{

−bn

[

(

R

Re

)1/n

− 1

]}

, (12)

where Ne is the projected number density at the effec-tive radius Re, n is the Sersic index, and the constantbn = 1.9992n − 0.3271 (Graham & Driver 2005). Wehave not fitted a background level because the radialcoverage of our data is not wide enough to estimate it.Based on Peng et al. (2011), the expected background ofpoint-like sources at HST/ACS resolution over this mag-nitude range is more than an order of magnitude belowthe number densities in our outermost bins (and morethan two orders of magnitude below the innermost bins),even when the GCs are split into blue and red groups.

NGC 4874 Globular Clusters 17

10-1 100

R [arcmin]101

102

103

NGC a

rcm

in−2

NGC 4873NGC 4871

PGC 44644NGC 4872

All I candidatesBlue+RedBlue peakRed peak

Figure 16. Number densities of GC candidates within the mag-nitude range 23.0 < I814 < 26.0 mag as a function of galactocen-tric radius R from NGC4874. The color-coded points representnumber density estimates within fixed 10′′ annuli for different GCsubsamples as described in the legend. The values are plotted atthe locations of the median galactocentric radius of the GCs withineach bin. The horizontal error bars represent the standard devia-tions of the radial positions within each annulus. The vertical errorbars indicate the Poisson errors on the counts, i.e.,

√N/(effective

area), within each annulus. The best-fit Sersic profiles for eachsubsample are displayed as color-coded curves. The labeled verti-cal dotted lines indicate the distance from NGC4874 to the fournearest surrounding galaxies. The half size of the apparent isopho-tal diameter D25 to µB = 25.0 mag/arcsec2 (de Vaucouleurs et al.1991) of NGC4874 is shown as a vertical dashed line.

The Sersic fits to the full radial ranges are shown inFigure 16; the reduced χ2 values for these fits are typ-ically ∼ 0.9, indicating that the fits provide reasonabledescriptions of the data over these radial ranges. For thefull color-selected sample of GCs with 0.5 < g475−I814 <1.6 (plotted as black points in Figure 16), we derive aSersic index nB+R = 1.5 ± 0.3 with an effective radiusof Re,B+R = 4.′2 ± 1.′5, corresponding to 122 ± 44 kpc.Peng et al. (2011), using the shallower ACSCCS data,but covering a larger area of the Coma cluster, foundn = 1.3±0.1, Re = 2.′2±0.′1, corresponding to 62±2 kpc.Our value of n agrees closely with this ACSCCS value,while our Re is larger by a factor of 2.0± 0.7, or a 1.4σdiscrepancy. Because our g475 imaging is significantlyless deep than I814, and could potentially affect the com-pleteness of innermost bins, we also fitted the numberdensities for all the I814-selected GC candidates over thesame 23.0 < I814 < 26.0 mag range, but without match-ing to the g475 detections. The resulting densities arerepresented by green points in Figure 16; as expected,they only differ at the 1-σ level for the innermost point.Our Sersic fit to this sample of “all” I814-selected GCsgives nall = 2.0± 0.4, Re,all = 3.′0± 0.′7, or 85± 21 kpc.For this case, Re agrees to better than 1.1σ, while n dif-fers by 1.7σ. Given the differences in depth and area forthese fits, the level of agreement with the ACSCCS studyis reasonable.Peng et al. (2011) chose not to fit the blue and red

GC components individually; this was in part becausethe separation between the two color components var-ied with position over the large area that they studied.For instance, they found that the blue peak of the GCpopulation within R < 50 kpc of NGC4874 occurredbetween the locations of the blue and red peaks in the

GC color distribution at larger radius. Since our deeperimaging data are limited to this one central pointing, wehere examine the radial distributions of the blue and redGCs separately, using the color at the local minimum(approximately g475−I814 = 1.0) of the nonparametrickernel density estimate shown in Figure 12 to divide theGCs into “blue” and “red” subpopulations. Fitting eachof these color components with Sersic profiles, we findnB = 1.9 ± 0.7, Re,B = 7.′0 ± 6.′3, nR = 1.2 ± 0.2,Re,R = 1.′6± 0.′2 for the blue and red GCs, respectively.The large uncertainty for the blue peak subpopulation

is mainly due to the fact that the effective radius is ap-parently much larger than our field of view. The large Re

for the blue GC distribution is likely related to the veryextended, mainly blue IGC population in Coma (Penget al. 2011). However, it is also related to the mainlyblue populations of GCs around the lower luminosity, butstill bright, elliptical galaxies in this field. Figure 16 in-dicates the radial locations of several neighboring galax-ies. The density of blue GCs appears to jump upwardnear the radii where NGC4871 and NGC4873 are lo-cated. While these galaxies are bright enough to harborsome red GCs, the mean color of these GCs would besignificantly bluer than those of NGC4874 (e.g., Penget al. 2006), meaning that some of the smaller neigh-bors’ red GCs would fall within the range of the blueGCs for NGC4874. Moreover, the reddest GCs in theseneighboring galaxies would be restricted to small galac-tocentric radii, where the completeness of our g475 datasuffers. The GCs at larger radii within these galaxies areoverwhelmingly blue.Because there is an inherent covariance between n and

Re for Sersic model fits when the measurements do notextend clearly beyond Re, we have refitted the variousGC samples with n fixed at 2.0. To avoid concern overpossible incompleteness near the bright galaxy center, wealso omit the central radial bin for these fits. With theseconstraints, we then find: Re,all = 4.′2 ± 0.′3, Re,B+R =4.′2± 0.′3, Re,B = 7.′2± 1.′2, and Re,R = 2.′6± 0.′2. Thus,when n is fixed and the central bin is omitted, matchingwith the g475 detections and limiting the color range doesnot change the resulting profile. However, we continueto find that the effective radius of the radial distributionof the blue GCs is significantly larger than that of thered GCs. Again, this is in part due to the contribution ofblue GCs from NGC4871, NGC4873, and other galaxies.It is also consistent with the radially declining fractionof red GCs found by Peng et al. (2011) over a larger areaof the Coma core, and many other studies that find thered GCs are more concentrated in giant ellipticals andwithin galaxy clusters (e.g., Faifer et al. 2011; Durrellet al. 2014).

4.5. Two-Dimensional Spatial Distribution

In order to investigate possible spatial differences be-tween the distributions of the stellar light and GCs inNGC4874, we constructed two-dimensional smoothedspatial number density maps of the GCs. For this pur-pose, we used GC candidates selected only from the I814photometry, i.e., the “All I candidates” sample in Fig-ure 16, since completeness may become a problem nearthe center of the galaxy for the ACS/F475W image.However, we have also repeated the full two-dimensionalanalysis using the matched F814W/F475W sample, and

18 Cho et al.

Figure 17. Example smoothed two-dimensional surface densitymap of the spatial distribution of GCs from the I-band photom-etry in the range 23.0 < I814 < 26.0 mag (color indicates surfacenumber density of GCs). The grid size of the two-dimensional his-togram is 50 pix with a Gaussian smoothing kernel of σ = 4 gridspacings for this example map; many other grid sizes and smooth-ings were explored. The white × symbol marks the center positionof NGC4874 from IRAF ELLIPSE fitting and the cyan point with er-ror bars marks the average center of the GC population. The smallwhite + symbols mark the locations of nine surrounding galaxies.

the results do not change in any significant way. Tocharacterize the two-dimensional GC distribution, we di-vided the ACS field into two-dimensional grids with var-ious grid sizes: 40, 50, 60, 70, 80, 90, and 100 pix on aside (recall the scale is 0.′′05 pix−1 for our ACS imaging)and calculated the number density of GCs within eachgrid cell. The resulting bi-dimensional histograms werethen smoothed with Gaussian kernels with varying stan-dard deviations of σ = 2, 3, and 4, in units of the gridspacing. Surprisingly, the peaks of these smoothed two-dimensional GC density distributions generally do notencompass the luminosity center of NGC 4874, i.e., theGCs in the inner region of this field have an off-centeredspatial distribution with respect to NGC4874. An ex-ample smoothed GC surface density map (50 pix gridsize with σ = 4 grid smoothing) is shown in Figure 17with the locations of the ten brightest galaxies marked.As evident in the figure, the peak of the GC density dis-tribution is displaced towards the south/southwest withrespect to the center of NGC4874.To quantify the centroid of the GC distribution, we

fitted elliptical isophotes (representing GC number iso-density contours) to the smoothed density maps usingthe IRAF ELLIPSE task. The distance from the lumi-nosity center of NGC 4874 to the center of each ellipseis plotted in Figure 18 as a function of the circularizedradius of each isophote rcir = a

√1− ǫ, where a is the

semi-major axis and ǫ is the ellipticity of each ellipse.We estimated the statistical significance of the centroidoffsets by bootstrap resampling of the GC spatial den-

Figure 18. Offset distance between the peak of NGC4874’sstarlight and the centers of elliptical isophotes of various circu-larized radii rcir. The elliptical isophotes (or isodensity contours)are fitted to multiple different 2-D representations (round points ofvarious colors indicating the different 2-D binnings and smoothingscales) of the GC number density distribution, as well as to thegalaxy light itself (black crosses). For the GC density distribution,the grid size (in pixels) of the spatial binning and the Gaussiansmoothing σ (in units of grid spacing) are indicated in the legend.The circularized radius is defined as rcir = a

√1−ǫ, where a is the

semi-major axis of each elliptical isophote (or isodensity contour)and ǫ is its ellipticity. Regardless of the factor-of-two range in gridsize and smoothing scale, we find that the central peak of the el-liptical model of the GC density distribution is offset by 3–5 kpc(6′′ to 10′′) with respect to the central peak of NGC4874 itself. Atlarge radii, beyond ∼ 10 kpc, the center of the elliptical GC isoden-sity contours approach to within about 1.5 kpc of the luminositycenter of NGC4874.

sity distribution 10,000 times before applying the two-dimensional smoothing. Figure 18 shows that, regard-less of the particular smoothing, the centroid of the GCdensity distribution is displaced from the galaxy centerby 4 ± 1 kpc (about 8′′) towards the south/southwestfrom the center of stellar light distribution. However, onlarger scales, rcir & 10 kpc, the centers of the GC isoden-sity contours approach within ∼ 1 kpc of the center ofthe galaxy isophotes. We note that Kim et al. (2013) alsoreported an offset (of ∼ 3 kpc) for the center of the GCsystem around NGC 1399, the cD galaxy in the Fornaxcluster.Most likely this displacement in the centroid of the

GC system is related to dynamical interactions withinthis very rich environment. We note that NGC4889,the brightest galaxy in the Coma cluster, is located ap-proximately 200 kpc to the east, and thus does notappear to be associated with the observed small offsetof NGC4874’s GC system. However, the offset doesalign closely with the direction towards NGC 4872, anS0 galaxy with a prominent bar. At a separation of only0.′82, or 24 kpc, NGC 4872 is the closest of the brightneighboring galaxies, and its velocity (from NED) differsby only 17± 4 km s−1 from NGC4874. Despite its lumi-nosity, NGC4872 does not have an obvious GC system ofits own (unlike NGC 4871 and NGC4873). It would beinteresting to explore through dynamical modeling if theobserved offset of the NGC4874 GC distribution couldbe related to dynamical interaction with NGC4872.

NGC 4874 Globular Clusters 19

Figure 19. 84′′ × 50′′ (∼ 40 kpc× 24 kpc) section near the top right corner of our deep F814W image (shown in full in Figure 1 (a)).The brightest galaxy here, near the lower left corner in this subimage, is the SB0 galaxy NGC4872 (labeled in Figure 1 (a)). The dwarfelliptical galaxy just to the upper right of center in this subimage is SDSS J125935.18+275605.0. It contains a sizable population of GCswith an asymmetric spatial distribution. Interestingly, this dE galaxy has a very high relative velocity of −2660± 160 km s−1 with respectto the cluster mean; this is about 2.5 times the cluster velocity dispersion. Several extremely diffuse galaxies, like those found in otherrecent studies, are also evident in this field.

4.6. A Dwarf Elliptical with an Asymmetrical GCSystem

In the course of our analysis of the galaxy light dis-tributions, we noticed one particular dwarf elliptical(dE) galaxy 1.′47 from NGC4874 that seemed rela-tively rich in GCs, but the GC distribution appearedstrikingly asymmetrical. Searching its coordinates inNED, we found that the galaxy was catalogued in theSloan Digital Sky Survey (SDSS) and is designatedSDSS J125935.18+275605.0; for convenience, we refer toit hereafter as SDSS J125935. This is the faintest of theten galaxies in our ACS images for which we performedisophotal modeling, and it lies near the top right cornerof our ACS field (outside our WFC3/IR imaging area);see the galaxy model panel in Figure 1. Figure 19 showsan 84′′ × 50′′ cutout of the region around this galaxyin our I814 image; the much brighter galaxy at lowerleft in this figure is NGC 4872, discussed above. Severalfainter, more diffuse, objects in this field appear simi-lar to the extremely diffuse galaxies first systematicallycatalogued in the Coma cluster by van Dokkum et al.(2015), and shown from deep Subaru imaging to be ubiq-uitous throughout the Coma cluster (Koda et al. 2015).The image also shows that the density of point sourcesaround the dE SDSS J125935 is not symmetric about thegalaxy’s center.Figure 20 further illustrates the spatial asymmetry of

the GC distribution in a 20′′ × 20′′ box around this dEby comparing it to seven “control” fields of the samesize and at the same radius with respect to the cD

galaxy NGC4874. In the figure, GCs in the matchedF475W+F814W sample with I814 < 26.0 mag are shownwith red circles, while those with 26 < I814 < 26.9 mag(i.e., down to the expected GCLF turnover) are shownwith smaller black circles. Although we have found thatthe completeness of the F475W detections is lower nearthe center of NGC4874 and the other bright ellipticals,the surface brightness of SDSS J125935 is low enoughthat completeness is not a serious issue to this mag-nitude. The dE itself has been subtracted using ourisophotal model. The green ellipse shows the r = 25mag arcsec−2 isophote from the SDSS, as reported byNED, and the dashed line marks the minor axis of thisellipse. The blue ellipse indicates the outermost isophotefor which we were able to constrain the galaxy’s elliptic-ity and position angle from our deep F814W image; ithas a semi-major axis of 8.′′25 and an ellipticity of 0.192.At the distance of Coma, this translates to semi-majorand semi-minor axes of 4.0 and 3.2 kpc, respectively.For the GCs with I814 < 26.0 mag, 9 of the 11 inside

the green ellipse lie to one side of the minor axis; theprobability of this occurring by chance is 6.5%, based onMonte Carlo tests. However, the asymmetry is not lim-ited to the brighter GCs; for those with I814 < 26.9 mag(red plus black circles), 17 of 23 lie to one side, whichhas a random probability of 3.5%. Considering the largerblue ellipse, 12 of 15 GCs with I814 < 26 mag, and 23of 31 GCs with I814 < 26.9 mag, lie to one side of theminor axis; these have random probabilities of 3.5% and1.1%, respectively. Thus, the asymmetry is significantwith ∼ 99% confidence. It is evident from Figure 20

20 Cho et al.

Figure 20. Spatial distributions of GCs around the dE galaxy SDSS J125935.18+275605.0 and in seven control fields at the same projecteddistance from the cD galaxy NGC4874. Each of the small fields is 20′′×20′′ in size, except the top center field, which is 20′′×15′′ because ofits proximity to the image edge. The dE has been subtracted via isophotal modeling; the larger (blue) ellipse marks the outermost isophotefor which we were able to constrain the galaxy’s ellipticity and position angle in our F814W imaging; it has semi-major and semi-minoraxes of 4.0 kpc and 3.2 kpc, respectively. For comparison, the green ellipse indicates the r = 25 mag arcsec−2 isophote catalogued in theSDSS; the dashed line marks the minor axis of this ellipse. Objects within red circles are GC candidates with I814 < 26 mag, while thosewithin black circles have 26.0 < I814 < 26.9 mag (see text). The GCs associated with SDSS J125935.18+275605.0 tend to fall towardsone side of the galaxy, in the direction opposite of NGC4874; the outer isophote also appears to be offset in this direction. Given this dEgalaxy’s very high velocity relative to Coma, its proximity to the cluster dynamical center, and its asymmetrically distributed GC system,it may be passing through the core of the cluster on its first infall and being stripped of much of its GC system.

that the outermost galaxy isophote is also offset slightly(centroid shift of 0.′′4) in the same direction as the GCs.We can estimate the size of the GC population in

SDSS J125935 from the 31± 5.6 GC candidates (the er-ror is based on Poisson statistics) with I814 < 26.9 magwithin the blue ellipse in Figure 20. This ellipse has anarea of 172.9 arcsec2, and based on the density of GCcandidates in the control fields, we would expect 14± 4contaminants (mostly GCs belonging to NGC4874) inthis area. The difference is 17 ± 7, which represents thenumber of GCs associated with SDSS J125935 brighterthan the GCLF turnover. For the total population,

assuming a symmetric GCLF, we double this numberto obtain 34 ± 14 GCs. To estimate the specific fre-quency SN (number per unit V luminosity; Harris &van den Bergh 1981), we measure the galaxy magni-tude within the same elliptical aperture for consistencyand find I814 = 17.46 ± 0.01 mag. The galaxy coloris g475−I814 = 1.072 ± 0.018 mag. Both of these val-ues are on the AB system and are corrected for Galac-tic extinction. Using empirical transformations fromBlakeslee et al. (2009, 2012), this color corresponds tog475−z850 ≈ 1.26 mag (typical of many dEs in the ACSVirgo Cluster Survey) and V−I ≈ 1.10 mag, where the

NGC 4874 Globular Clusters 21

latter value is on the standard Vega-based system. To getthe absolute V magnitude, we subtract 0.42 mag fromI814 to convert to the Vega system, then subtract thedistance modulus (m−M) = 35.0, and finally add theestimated V−I to obtain MV = −16.86 mag. The spe-cific frequency is then SN = 6.1± 2.6 (the error includesan estimated 10% uncertainty on the galaxy luminosity).While this SN would be above average for a large galaxy,it is well within the range for dEs of similar luminosityin the Virgo cluster (Peng et al. 2008).Remarkably, SDSS J125935 has a heliocentric radial

velocity of 4193±154 km s−1, measured by Biviano et al.(1995). This is nearly 3000 km s−1 less than the velocityof 7176±3 km s−1 for NGC4874 (Trager et al. 2008). Ac-cording to Colless & Dunn (1996), the main componentof the Coma cluster centered on NGC4874 has a meanvelocity 〈v〉 = 6853±54 km s−1 and line-of-sight velocitydispersion σComa = 1082 km s−1. Thus, SDSS J125935has a relative velocity of −2660 ± 160 km s−1 with re-spective to the cluster mean, or −(2.5± 0.2)σComa. Nu-merical simulations indicate that dEs at such small clus-tercentric radii and high relative velocities are likely tobe on their first infall into the cluster core (Smith et al.2013, 2015). The same simulations show that dwarfs thatpass through the cluster centers can lose a large fraction,even the majority, of their GC systems (see also Aguilar& White 1986 for illustrations of how similar encounterscan result in asymmetric distributions of GC-like testmasses).We suggest that SDSS J125935 is a dwarf elliptical

with a relatively rich GC system, similar to some dEsin Virgo, that has recently fallen at high velocity intothe core of the Coma cluster and is undergoing strippingof its GC system. Unfortunately, it is presently unfeasi-ble to measure spectroscopic velocities for the surround-ing point sources to determine what fraction belong toSDSS J125935. This would be another interesting systemfor detailed dynamical modeling.

5. SUMMARY & CONCLUSIONS

We have studied the rich GC system of NGC4874,the cD galaxy in the core of the Coma cluster of galax-ies, using optical HST g475 and I814 imaging from theACS/WFC and near-IR H160 imaging from WFC3/IR.The GC system of NGC4874 and the surrounding Comacore was previously studied in g475 and I814 as part of theACSCCS (Peng et al. 2011), and we find excellent photo-metric agreement with that study, but the exposure timeof our I814 observations is more than seven times that ofthe ACSCCS imaging, giving a limiting magnitude morethan a magnitude fainter in this bandpass. Because weadded the ACSCCS g475 observations to our own, thestacked g475 image has a factor of two more exposuretime than the ACSCCS in the overlap region; tests showthat our g475−I814 color measurements have a factor oftwo smaller errors than those from the ACSCCS. In ad-dition, we include new deep F160W observations, withan exposure time slightly longer than that of I814, overthe smaller field of the WFC3/IR.Because the I814−H160 color for old stellar populations

measures red giant branch temperature, it should be sen-sitive mainly to metallicity, while g475−I814 also dependson horizontal branch morphology and the location of

the main sequence turnoff. Over most of the luminos-ity range probed by our data, there exists clear bimodal-ity in the distribution of g475−I814 colors of our selectedGC candidates. This optical bimodality can be traced atleast to I814 = 26 mag, corresponding to M814 = −9 magat the distance of Coma, or MV ≈ −8 mag for typicalGCs. From a Gaussian mixture modeling analysis as afunction of magnitude, we find that at the brightest mag-nitudes, the blue peak exhibits a very strong “tilt” to-wards redder colors, with a slope d(g475−I814)/dI814 =−0.082 ± 0.020 for I814 < 25 mag. Based on the em-pirical calibration of metallicity as a function of photo-metric color from the ACS Virgo Cluster Survey, thiscorresponds to a very steep mass-metallicity scaling ofZ ∝ M1.4±0.4

GC at these highest masses.The GMM analysis for the I814−H160 color distri-

bution is generally less robust than for g475−I814, es-pecially when the sample is further broken down bymagnitude. We therefore instead examined the varia-tion in the overall mean I814−H160 color in the samemagnitude range as for g475−I814. Again for I814 <25 mag, we find a steep slope in the mean I814−H160

color of d(I814−H160)/dI814 = −0.093 ± 0.013. Al-though there is no empirical relation between I814−H160

and metallicity for GCs, the linear approximation tothe relation between I814−H160 and g475−I814 givesd(I814−H160)/d(g475−I814) ≈ 1.1, which again impliesZ ∝ M1.4

GC. Thus, the mean metallicity scaling derivedfrom the full I814−H160 color range is the same as thatfound from the blue component of the g475−I814 colordistribution. However, the color-magnitude tilt is nota simple linear relation, and if we extend the linear fitanother magnitude fainter to I814 = 26 mag, then thebest-fit slopes are roughly a factor of three shallower,giving scalings of Z ∝ M0.5±0.2

GC , consistent with the typ-ical scaling found by Mieske et al. (2010) over a similarmass range.As a consequence of the tilted color-magnitude rela-

tions, the color distributions change as a function ofmagnitude. Both the g475−I814 and I814−H160 distribu-tions appear broad and red, with no evidence for multiplepeaks for the brightest GCs at I814 < 23 mag. Fainterthan this, g475−I814 is clearly bimodal, with the promi-nence of the red peak decreasing at progressively faintermagnitudes. The bimodality is less evident in I814−H160,but the same general trend occurs, with the histogramtransitioning from a redward tilt to being skewed towardsthe blue at fainter magnitudes. Because of the blue tiltat bright magnitudes and increased measurement errorat faint magnitudes, the bimodality is most evident for23 < I814 < 25 mag, and we have compared the GMMbimodal decompositions for g475−I814 and I814−H160 forthe identical sample of GC candidates over this magni-tude range. Once the four bluest objects in I814−H160

are excluded, the red:blue decompositions are consistent,with red fractions f2 of 0.61 ± 0.08 for g475−I814 and0.54± 0.13 for I814−H160.While the separation of the peaks in units of the peak

dispersion is very clear in g475−I814 with D = 2.88±0.28,it is less clear in I814−H160 with D = 2.23 ± 0.26,even though the separation in magnitudes is essentiallyidentical. The reason for this is that the blue peak ismuch narrower in g475−I814, with a dispersion σ1 =

22 Cho et al.

0.067± 0.014 mag, compared to σ1 = 0.098± 0.013 magfor I814−H160, a difference of nearly 50%. For the redpeaks, the dispersions are σ2 = 0.105 ± 0.014 mag andσ2 = 0.126 ± 0.017 mag for g475−I814 and I814−H160,respectively. Previous studies of optical GC color distri-butions (e.g., Peng et al. 2006, 2009; Harris et al. 2016)also found that the blue peak was significantly narrowerthan the red peak; however, Peng et al. (2006) pointedout that the dispersion in [Fe/H] was actually larger forthe blue peak because of the steeper variation in metal-licity with color for the blue component of the GCs.The differences in the blue and red color dispersions forg475−I814 as compared to I814−H160 suggests that thecolors follow different color-metallicity relations, despitetheir nearly identical total range in color. In particu-lar, the metallicity slopes at blue and red colors mustbe more similar (i.e., weaker nonlinearity) for I814−H160

than for g475−I814. Consistent with this, we find thatthe variation in I814−H160 with g475−I814 is nonlinear,with an inflected shape that can be described well by acubic polynomial.We have compared the radial distributions of the blue

and red GCs over the wider ACS field of view. Consis-tent with previous studies, we find that the blue GCsfollow a more spatially extended radial profile than thered GCs. Interestingly, for this field located in the densecentral region of the rich Coma cluster of galaxies, thebroader extent of the blue GCs is at least partially theresult of the GCs associated with the fainter neighboringearly-type cluster galaxies, whose GC systems are pre-dominantly blue, especially at large galactocentric radii.This is consistent with the view that a significant frac-tion of the blue GCs in the halos of massive galaxies areadded through the accretion or stripping of lower lumi-nosity satellite galaxies.Curiously, the center of the spatial distribution of the

GCs in this field is offset by 4 ± 1 kpc from the cen-ter of NGC 4874 itself. This offset does not appear toresult from the superposition of the GC population ofany neighboring galaxy, but it is likely the signatureof past dynamical interaction. The most likely candi-date for this is NGC 4872, a bright SB0 galaxy 24 kpcfrom the center of NGC 4874 with a velocity differenceof less than 20 km s−1. Although NGC4872 does nothave a significant GC population of its own, the 4 kpcdisplacement in the centroid of the NGC4874 GC sys-tem lies along the line towards NGC4872. We have alsodiscussed the asymmetry of the GC system of the dEgalaxy SDSS J125935, which is projected 42 kpc fromNGC4874, but has a relative velocity of −2983 km s−1

with respect to the cD, and −2660 km s−1 with respectto the cluster mean. The dE has a specific frequencySN = 6.1 ± 2.6. The likelihood of the asymmetry in itsGCs occurring by chance is ∼ 1%. We suggest that thisdE is on an initial high-velocity infall into the cluster coreand its GC system is in the process of being stripped.Interestingly, based on stellar absorption line indices,

Trager et al. (2008) concluded that NGC4874 andneighboring early-type galaxies showed evidence for anintermediate-age stellar population component, whichwould imply a significant star formation event severalbillion years ago. For now, it remains a matter of specu-lation whether this proposed star formation event in therelatively recent past is associated with the spatial offset

of the NGC4874 GCs. It is also unknown whether or notsuch an event may have produced any significant popu-lation of intermediate-age GCs. If so, one would expectthe color-metallicity and color-color relations in this fieldto differ from those in massive galaxies with exclusivelyold GC populations, as predicted from stellar popula-tion models (Yoon & Chung 2009). Usher et al. (2015)have shown that the color-metallicity relations do indeedvary among early-type galaxies, and that this variationappears to correlate with galaxy luminosity and color;further work is needed to understand the detailed causesof these variations.We are currently carrying out an optical-NIR photo-

metric study of GCs in a much larger set of 16 early-type galaxies in the Fornax and Virgo clusters by cross-matching our HST WFC3/IR data (Jensen et al. 2015)with the published F475W and F850LP catalogs fromthe ACS Fornax and Virgo Cluster Surveys (Jordan et al.2009, 2015). Because these galaxies cover a large rangein luminosity and color, this sample will shed light onwhether optical/NIR color-color relations show varia-tions with galaxy type similar to those found by Usheret al. (2015) for the relation between optical color andmetallicity estimated from the CaT index, as well as illu-minating differences in the ways that different broadbandcolors trace the underlying metallicity. Unfortunately,existing samples of spectroscopically estimated metallic-ities for massive early-type galaxies are of inhomogeneousquality, tend to be based on a small number of metal ab-sorption line indices, and often have large uncertaintiesin excess of 0.5 dex. A large sample of uniformly high-quality spectroscopic metallicities (σ[Fe/H] ∼ 0.1 dex)and ages determined over a broad spectral range for hun-dreds of GCs spanning the full color range in a nearbycD galaxy (which likely combines GCs from a diversemix of other cluster galaxies) would be an invaluable re-source for the community. Such a sample would allow usto calibrate empirically the detailed forms of the color-metallicity relations from the UV to the NIR, and thusconstrain the enrichment histories of more distant galax-ies from photometric studies alone; it would also enablecrucial tests of the stellar population models. The evolu-tionary histories of massive galaxies and their surround-ing environments are encoded in the properties of theancient systems of GCs that surround them; decodingthese histories remains a major ongoing archaeologicaleffort in extragalactic astronomy.

Support for this work was provided by the Na-tional Research Foundation of Korea to the Center forGalaxy Evolution Research (CGER). H.C. thanks theNational Research Council of Canada’s Herzberg Astron-omy & Astrophysics for hospitality during several visits.J.P.B. thanks the CGER at Yonsei University for hos-pitality on numerous occasions. We thank Pat Cote,Ruben Sanchez-Janssen, Laura Sales, and Rory Smithfor helpful conversations. This research has made use ofthe NASA/IPAC Extragalactic Database (NED) whichis operated by the Jet Propulsion Laboratory, CaliforniaInstitute of Technology, under contract with the NationalAeronautics and Space Administration.Facility: HST (WFC3/IR, ACS/WFC)

NGC 4874 Globular Clusters 23

REFERENCES

Aguilar, L. A., & White, S. D. M. 1986, ApJ, 307, 97Alamo-Martınez, K. A., Blakeslee, J. P., Jee, M. J., et al. 2013,

ApJ, 775, 20Anderson, J., & Bedin, L. R. 2010, PASP, 122, 1035Bailin, J., & Harris, W. E. 2009, ApJ, 695, 1082Beasley, M. A., Bridges, T., Peng, E., et al. 2008, MNRAS, 386,

1443Bekki, K., & Yahagi, H. 2006, MNRAS, 372, 1019Bekki, K., Yahagi, H., Nagashima, M., & Forbes, D. A. 2008,

MNRAS, 387, 1131Benıtez, N. 2000, ApJ, 536, 571Bergbusch, P. A., & VandenBerg, D. A. 2001, ApJ, 556, 322Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393Biviano, A., Durret, F., Gerbal, D., et al. 1995, A&AS, 111, 265Blakeslee, J. P. 1999, AJ, 118, 1506Blakeslee, J. P., Anderson, K. R., Meurer, G. R., et al. 2003,

Astronomical Data Analysis Software and Systems XII, 295,257

Blakeslee, J. P., Cho, H., Peng, E. W., et al. 2012, ApJ, 746, 88Blakeslee, J. P., Jordan, A., Mei, S., et al. 2009, ApJ, 694, 556Blakeslee, J. P., & Tonry, J. L. 1995, ApJ, 442, 579Blakeslee, J. P., Tonry, J. L., & Metzger, M. R. 1997, AJ, 114, 482Bohlin, R. C. 2012, Instrument Science Report ACS 2012-01

(Baltimore, MD: STScI)Brodie, J. P., Romanowsky, A. J., Strader, J., et al. 2014, ApJ,

796, 52Brodie, J. P., & Strader, J. 2006, ARA&A, 44, 193Brodie, J. P., Usher, C., Conroy, C., et al. 2012, ApJ, 759, 33Bruzual, G., & Charlot, S. 2003, MNRAS, 344, 1000Busko, I. C. 1996, in ASP Conf. Ser. 101, Astronomical Data

Analysis Software and Systems V, ed. G. H. Jacoby & J.Barnes (San Francisco, CA: ASP), 139

Cantiello, M., & Blakeslee, J. P. 2007, ApJ, 669, 982Cantiello, M., Blakeslee, J. P., Raimondo, G., et al. 2014, A&A,

564, L3Carter, D., Goudfrooij, P., Mobasher, B., et al. 2008, ApJS, 176,

424Chiboucas, K., Tully, R. B., Marzke, R. O., et al. 2011, ApJ, 737,

86Chies-Santos, A. L., Larsen, S. S., Cantiello, M., et al. 2012,

A&A, 539, A54Colless, M., & Dunn, A. M. 1996, ApJ, 458, 435Cote, P., Blakeslee, J. P., Ferrarese, L., et al. 2004, ApJS, 153, 223Dirsch, B., Richtler, T., Geisler, D., et al. 2003, AJ, 125, 1908Dotter, A., Chaboyer, B., Jevremovic, D., et al. 2007, AJ, 134, 376Dotter, A., Sarajedini, A., Anderson, J., et al. 2010, ApJ, 708, 698Dressler, A. 1980, ApJ, 236, 351Durrell, P. R., Cote, P., Peng, E. W., et al. 2014, ApJ, 794, 103Faifer, F. R., Forte, J. C., Norris, M. A., et al. 2011, MNRAS,

416, 155Forte, J. C., Vega, E. I., Faifer, F. R., et al. 2014, MNRAS, 441,

1391Fruchter, A., Sosey, M., Hack, W., et al. 2009, The MultiDrizzle

Handbook, Version 3.0 (Baltimore, MD: STScI)Graham, A. W. & Driver, S. P. 2005, PASA, 22, 118Harris, W. E. 1987, ApJL, 315, L29Harris, W. E. 2009, ApJ, 699, 254Harris, W. E., Blakeslee, J. P., Whitmore, B. C., et al. 2016, ApJ,

817, 58Harris, W. E., Harris, G. L. H., & Alessi, M. 2013, ApJ, 772, 82Harris, W. E., Kavelaars, J. J., Hanes, D. A., et al. 2009, AJ, 137,

3314Harris, W. E., & van den Bergh, S. 1981, AJ, 86, 1627Harris, W. E., Whitmore, B. C., Karakla, D., et al. 2006, ApJ,

636, 90Hudson, M. J., Harris, G. L., & Harris, W. E. 2014, ApJL, 787, L5Jedrzejewski, R. 1987, MNRAS, 226, 747Jee, M. J., Blakeslee, J. P., Sirianni, M., et al. 2007, PASP, 119,

1403Jensen, J. B., Blakeslee, J. P., Gibson, Z., et al. 2015, ApJ, 808, 91Jordan, A., Blakeslee, J. P., Cote, P., et al. 2007, ApJS, 169, 213Jordan, A., Blakeslee, J. P., Peng, E. W., et al. 2004, ApJS, 154,

509Jordan, A., Cote, P., Blakeslee, J. P., et al. 2005, ApJ, 634, 1002Jordan, A., Peng, E. W., Blakeslee, J. P., et al. 2009, ApJS, 180,

54

Jordan, A., Peng, E. W., Blakeslee, J. P., et al. 2015, ApJS, 221,13

Keller, S. C., Mackey, D., & Da Costa, G. S. 2012, ApJ, 744, 57Kim, H.-S., Yoon, S.-J., Sohn, S. T., et al. 2013, ApJ, 763, 40Koda, J., Yagi, M., Yamanoi, H., & Komiyama, Y. 2015, ApJL,

807, L2

Koekemoer, A. M., Fruchter, A. S., Hook, R. N., & Hack, W.2003, in The 2002 HST Calibration Workshop, ed. S. Arribas,A. M. Koekemoer, & B. Whitmore (Baltimore, MD: STScI),337

Kubo, J. M., Stebbins, A., Annis, J., et al. 2007, ApJ, 671, 1466Lee, Y.-W., Demarque, P., & Zinn, R. 1994, ApJ, 423, 248Lee, M. G., Park, H. S., & Hwang, H. S. 2010, Science, 328, 334Liu, C., Peng, E. W., Cote, P., et al. 2015, ApJ, 812, 34Mackey, A. D., Huxor, A. P., Ferguson, A. M. N., et al. 2010,

ApJL, 717, L11Madrid, J. P., Graham, A. W., Harris, W. E., et al. 2010, ApJ,

722, 1707Marın-Franch, A. & Aparicio, A. 2002, ApJ, 568, 174Masters, K. L., Jordan, A., Cote, P., et al. 2010, ApJ, 715, 1419Mieske, S., Jordan, A., Cote, P., et al. 2006, ApJ, 653, 193Mieske, S., Jordan, A., Cote, P., et al. 2010, ApJ, 710, 1672Mistani, P. A., Sales, L. V., Pillepich, A., et al. 2016, MNRAS,

455, 2323Moore, B., Diemand, J., Madau, P., et al. 2006, MNRAS, 368, 563Muratov, A. L., & Gnedin, O. Y. 2010, ApJ, 718, 1266Ostrov, P. G., Forte, J. C., & Geisler, D. 1998, AJ, 116, 2854Peng, E. W., Ferguson, H. C., Goudfrooij, P., et al. 2011, ApJ,

730, 23Peng, E. W., Jordan, A., Blakeslee, J. P., et al. 2009, ApJ, 703, 42Peng, E. W., Jordan, A., Cote, P., et al. 2006, ApJ, 639, 95Peng, E. W., Jordan, A., Cote, P., et al. 2008, ApJ, 681, 197Puzia, T. H., Paolillo, M., Goudfrooij, P., et al. 2014, ApJ, 786, 78Puzia, T. H., Zepf, S. E., Kissler-Patig, M., et al. 2002, A&A,

391, 453Richtler, T. 2006, BASI, 34, 83Schlafly, E. F. & Finkbeiner, D. P. 2011, ApJ, 737, 103Schlegel, D. J., Finkbeiner, D. P., & Davis, M. 1998, ApJ, 500,

525Sersic, J. L. 1963, BAAA, 6, 41Sirianni, M., Jee, M. J., Bentez, N., et al. 2005, PASP, 117, 1049Smith, R., Sanchez-Janssen, R., Beasley, M. A., et al. 2015,

MNRAS, 454, 2502Smith, R., Sanchez-Janssen, R., Fellhauer, M., et al. 2013,

MNRAS, 429, 1066Spitler, L. R., & Forbes, D. A. 2009, MNRAS, 392, L1Strader, J., Brodie, J. P., Spitler, L., & Beasley, M. A. 2006, AJ,

132, 2333Strader, J., & Smith, G. H. 2008, AJ, 136, 1828Trager, S. C., Faber, S. M., & Dressler, A. 2008, MNRAS, 386,

715Usher, C., Forbes, D. A., Brodie, J. P., et al. 2012, MNRAS, 426,

1475Usher, C., Forbes, D. A., Brodie, J. P., et al. 2015, MNRAS, 446,

369van Dokkum, P. G., Abraham, R., Merritt, A., et al. 2015, ApJL,

798, L45Vanderbeke, J., West, M. J., De Propris, R., et al. 2014, MNRAS,

437, 1734Wehner, E. M. H., Harris, W. E., Whitmore, B. C., et al. 2008,

ApJ, 681, 1233West, M. J., Jordan, A., Blakeslee, J. P., et al. 2011, A&A, 528,

A115Woodley, K. A., Harris, W. E., Puzia, T. H., et al. 2010, ApJ,

708, 1335Yoon, S.-J. & Chung, C. 2009, in ESO Astrophysics Symposia,

Globular Clusters – Guides to Galaxies, ed. T. Richtler & S.Larsen (Berlin: Springer), 381

Yoon, S.-J., Lee, S.-Y., Blakeslee, J. P., et al. 2011a, ApJ, 743, 150Yoon, S.-J., Sohn, S. T., Lee, S.-Y., et al. 2011b, ApJ, 743, 149Yoon, S.-J., Yi, S. K., & Lee, Y.-W. 2006, Science, 311, 1129Zaritsky, D., Colucci, J. E., Pessev, P. M., et al. 2014, ApJ, 796,

71

24 Cho et al.

Table 2The GMM Analysis Results for the Color Distributions Shown in Figures 10, 11, and 12 (full sample)

I814 Ntot µ1 σ1 µ2 σ2 f2 D kurt p(χ2)(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

g475−I814 of ACS F814W and F475W matched samples

21.5–23.0 26 1.065±0.030 0.048±0.017 1.219±0.028 0.053±0.016 0.566±0.167 3.07±0.72 −1.105 0.56623.0–24.0 164 0.922±0.035 0.078±0.021 1.154±0.026 0.091±0.015 0.634±0.126 2.74±0.40 −0.756 0.00324.0–25.0 775 0.866±0.011 0.068±0.010 1.112±0.022 0.127±0.012 0.627±0.071 2.42±0.26 −0.736 <0.00125.0–26.0 1798 0.839±0.010 0.071±0.014 1.090±0.034 0.190±0.017 0.702±0.092 1.75±0.28 −0.557 <0.01023.0–25.0a 939 0.873±0.010 0.071±0.008 1.123±0.017 0.119±0.010 0.624±0.055 2.55±0.21 −0.775 <0.001

I814−H160 of ACS F814W and WFC3/IR F160W matched samples

21.5–23.0 12 0.638±0.036 0.102±0.028 0.914±0.050 0.026±0.019 0.161±0.129 3.71±1.23 −0.854 0.27523.0–24.0 74 0.400±0.032 0.071±0.018 0.633±0.032 0.088±0.016 0.603±0.119 2.92±0.49 −0.899 0.08824.0–25.0 319 0.397±0.044 0.138±0.024 0.675±0.078 0.110±0.031 0.271±0.184 2.24±0.41 −0.483 0.05025.0–26.0 643 0.419±0.167 0.202±0.040 0.769±0.111 0.138±0.054 0.152±0.240 2.02±0.71 −0.111 0.19023.0–25.0a 393 0.399±0.035 0.133±0.020 0.660±0.052 0.108±0.020 0.330±0.144 2.17±0.32 −0.499 0.030

Note. — Column lists: (1) I814 AB magnitude range; (2) total number of objects in the analyzed sample; (3) mean anduncertainty of the first mode in the double Gaussian mixture model; (4) standard deviation and uncertainty of the first mode;(5) mean and uncertainty of the second mode; (6) standard deviation and uncertainty of the second mode; (7) fraction ofobjects assigned to the second component of the double Gaussian mixture model; (8) separation between the peaks relative totheir Gaussian σ; (9) kurtosis of the distribution; (10) likelihood that the sample was drawn from a single Gaussian distribution.a The results for the color distributions over the 23.0 < I814 < 25.0 magnitude range shown in Figure 12.

Table 3

The GMM Analysis Results for the Color Distributions Shown in Figure 13 (matched subsample)

Case Ntot µ1 σ1 µ2 σ2 f2 D kurt p(χ2)(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

g475−I814 of ACS F814W + F475W + WFC3/IR F160W matched samples

σ1 6= σ2 392 0.884±0.016 0.067±0.015 1.139±0.026 0.106±0.016 0.604±0.083 2.89±0.32 −0.986 <0.001σ1 = σ2 392 0.911±0.009 0.087±0.004 1.168±0.009 0.087±0.004 0.493±0.031 2.94±0.20 −0.986 <0.001

I814−H160 of ACS F814W + F475W + WFC3/IR F160W matched samples

σ1 6= σ2 392 0.399±0.031 0.133±0.018 0.663±0.047 0.107±0.018 0.324±0.128 2.18±0.29 −0.501 0.034σ1 = σ2 392 0.373±0.015 0.121±0.008 0.629±0.019 0.121±0.008 0.436±0.054 2.12±0.28 −0.501 0.007

Note. — Column (1) shows whether the double Gaussian mixture is for the heteroscedastic or homoscedastic case.Columns (2)-(10) are the same as in Table 2.

Table 4

Same as Table 3 but in different color ranges.

I814−H160 Ntot µ1 σ1 µ2 σ2 f2 D kurt p(χ2)(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

g475−I814 of ACS F814W + F475W + WFC3/IR F160W matched samples

−0.022 to 0.877 392 0.884±0.016 0.067±0.015 1.139±0.026 0.106±0.016 0.604±0.083 2.89±0.32 −0.986 <0.0010.059 to 0.877 390 0.885±0.018 0.067±0.016 1.140±0.026 0.105±0.016 0.604±0.087 2.88±0.30 −0.980 <0.0010.133 to 0.877 388 0.885±0.016 0.067±0.014 1.139±0.023 0.105±0.014 0.609±0.082 2.88±0.28 −0.978 <0.001

I814−H160 of ACS F814W + F475W + WFC3/IR F160W matched samples

−0.022 to 0.877 392 0.399±0.031 0.133±0.018 0.663±0.047 0.107±0.018 0.324±0.128 2.18±0.29 −0.501 0.0340.059 to 0.877 390 0.368±0.033 0.111±0.015 0.627±0.042 0.119±0.017 0.461±0.132 2.25±0.26 −0.667 0.0020.133 to 0.877 388 0.353±0.031 0.098±0.013 0.606±0.041 0.126±0.017 0.538±0.126 2.23±0.26 −0.727 <0.001

Note. — Column (1) shows minimum and maximum values of I814−H160 color for the objects in the sample (the g475−I814ranges are not modified). Columns (2)-(10) are the same as in Table 2.